Title: Pt film grain border detection

author: Nikolay Chehlarov

email: chehlarow@yahoo.com

git: https://github.com/Chehlarov

date: 17.02.2022.

Abstract

This work describes the development of an algorithm for image segmentation. The goal is to detect Pt grain borders from SEM images. Four U-nets are implemented and evaluated. The best U-net type for the task proved to be Spatial Attention U-Net. Hyper parameter tuning is performed to find optimal settings for the model.

Problem definition

Background

Pt films are used in engineering for various purposes. For example for temperature or strain sensing. The distribution of grain sizes is an important property for the product function. Traditional based approaches failed to achieve acceptable results. ML algorithm was developed with good performance (https://github.com/Chehlarov/Machine-Learning/tree/main/00%20-%20project). In production environment photos with different noise level and scale have to be analyzed. The ML approach did not deliver the expected generalization and production readiness. Deep learning approach is expected to meet the production needs.

Goal

Create a model to perform image segmentation of Pt grains from a SEM image. Pixels should be classified as border or not border.

Architecture selection

The task is semantic segmentation, several suitable architecture exist. The difference between grain border segmentation and typical segmentation task is that grain borders are continues curves with small width. Similar standard task is Retinal Vessel Segmentation . Looking into the top performing models, one could easily notice that several models are based on U-net architecture. U-net is also the most implemented paper. U-net types can summarized to:

• U-net (classical)
• R2U-Net (Recurrent Residual)
• SA-UNet (Spatial Attention U-Net; top performer among U-nets)
• DUNet (deformable network)
• BCDU-Net (Bi-Directional ConvLSTM U-Net)
• Residual U-Net
• Att U-Net (Attention U-Net)
• Attention Residual U-Net
• Channel Attention Residual U-Net (CAR U-net)

U-net is supposed to perform well even on small dataset.The available labeled dataset for border segmentation is small and U-net might be suitable solution. A couple of the U-net versions will be implemented and compared:

• Spatial Attention U-Net
• U-net
• Attention U-net
• Attention Residual U-net

Spatial Attention U-Net

Highlights:

• works well with small datasets
• uses attention module which infers the attention map along the spatial dimension
• employs structured dropout convolutional blocks instead of the original convolutional blocks (prevents overfitting)
• top results achieved on retinal blood vessels dataset

SA U-net architecture attention module

U-Net

Highlights:

• very popular, proven model for image segmentation
• the base model used for the four implemented models in this work
• originally developed for biomedical image segmentation
• The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization

U-net architecture

Attention U-Net

Highlights:

• based on U-net
• adds attention gates to the skip connections
• learn to suppress irrelevant regions
• minimal computational overhead

Attention U-net architecture

Attention gate

Attention Residual U-net

Highlights:

• based on U-net
• combines residual U-net skip connections and attention U-net gates
• overcomes problems with vanishing gradients

residual block

Implementation

The implementation is based on several Github repositories. There are modifications to adapt to the current task. The code is based on tensorflow 2.

Imports

import tensorflow as tf
from tensorflow.keras import models, layers, regularizers
from tensorflow.keras import backend as K
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.preprocessing.image import ImageDataGenerator

import keras_tuner as kt

import tensorflow_addons as tfa

from focal_loss import BinaryFocalLoss

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.animation as animation

from skimage import io, color, measure
from skimage.util import img_as_float32, img_as_ubyte
from skimage import filters
from skimage.feature import peak_local_max
from skimage.morphology import binary_dilation, flood_fill #skeletonize, disk, binary_closing, 
from skimage.segmentation import clear_border, watershed

from scipy import ndimage

import pandas as pd

import cv2

import time

import os

# import seaborn as sns

Function for models building and evaluation

This section contains the main building blocks for the U-net models.

# A few metrics and losses

def dice_coef(y_true, y_pred):
    """
    Basically F1 score; harmonic mean between precition and recall
    https://www.youtube.com/watch?v=AZr64OxshLo&t=15s
    better for optimization (loss function) because it tries to balance precition and recall
    """
    y_true_f = K.flatten(y_true)
    y_true_f = tf.cast(y_true_f, tf.float32)
    y_pred_f = K.flatten(y_pred)
    intersection = K.sum(y_true_f * y_pred_f)
    return (2.0 * intersection + 1.0) / (K.sum(y_true_f) + K.sum(y_pred_f) + 1.0)


def jacard_coef(y_true, y_pred):
    """
    bassically IoU
    https://www.youtube.com/watch?v=AZr64OxshLo&t=15s
    better as metric because it is easy to interprit
    """
    y_true_f = K.flatten(y_true)
    y_true_f = tf.cast(y_true_f, tf.float32)
    y_pred_f = K.flatten(y_pred)
    intersection = K.sum(y_true_f * y_pred_f)
    return (intersection + 1.0) / (K.sum(y_true_f) + K.sum(y_pred_f) - intersection + 1.0)

# not used
# def jacard_coef_loss(y_true, y_pred):
#     return 1 - jacard_coef(y_true, y_pred)


def dice_coef_loss(y_true, y_pred):
    return 1 - dice_coef(y_true, y_pred)


def mae_euclidean(y_true, y_pred):
    """
    Mean square error of euclean distance between the predicted distance map and the true distance map
    This metric is not very stable, one false positive pixel inside a grain can change the metric significantly.
    The metric has bias towards the grain size distribution - fine grain structures will have lower 
    euclidean distances and hence smaller MSE values.
    this function is not differentiable and cannot be used as loss function.
    """
    y_true_inv = (y_true == 0)
    y_true_inv = tf.cast(y_true_inv, tf.uint8)
    y_true_distance_map = tfa.image.euclidean_dist_transform(y_true_inv)  
    
    y_pred_inv = (y_pred < 0.5)
    y_pred_inv = tf.cast(y_pred_inv, tf.uint8)
    y_pred_distance_map = tfa.image.euclidean_dist_transform(y_pred_inv)
    
    # res = tf.math.squared_difference(y_pred_distance_map, y_true_distance_map) # squared error leads to high penaulty for misclasification
    res = tf.math.abs(tf.math.subtract(y_pred_distance_map, y_true_distance_map))
    res = tf.keras.backend.mean(res)
    
    return res

#unit test
y_true = np.array([0,  0, 1, 0, 1, 0, 1])
y_pred = np.array([1., 1, 0, 0, 1, 0, 1])
y_true = tf.expand_dims(y_true, -1)
y_pred = tf.expand_dims(y_pred, -1)
assert(np.abs(mae_euclidean(y_true, y_pred).numpy() - 0.5714286) < 0.0001)

# main blocks of the U-nets

class DropBlock2D(layers.Layer):
    """
    See: https://arxiv.org/pdf/1810.12890.pdf
    Used in SA-UNet
    """
    
    def _normalize_data_format(self, value):
        # https://stackoverflow.com/questions/53442190/importerror-cannot-import-name-normalize-data-format
        if value is None:
            value = K.image_data_format()
        data_format = value.lower()
        if data_format not in {'channels_first', 'channels_last'}:
            raise ValueError('The `data_format` argument must be one of '
                             '"channels_first", "channels_last". Received: ' +
                             str(value))
        return data_format
    
    def __init__(self,
                 block_size,
                 keep_prob,
                 sync_channels=False,
                 data_format=None,
                 **kwargs):
        """Initialize the layer.
        :param block_size: Size for each mask block.
        :param keep_prob: Probability of keeping the original feature.
        :param sync_channels: Whether to use the same dropout for all channels.
        :param data_format: 'channels_first' or 'channels_last' (default).
        :param kwargs: Arguments for parent class.
        """
        super(DropBlock2D, self).__init__(**kwargs)
        self.block_size = block_size
        self.keep_prob = keep_prob
        self.sync_channels = sync_channels
        self.data_format = self._normalize_data_format(data_format) 
        self.input_spec = tf.keras.layers.InputSpec(ndim=4)
        self.supports_masking = True
        


    def get_config(self):
        config = {'block_size': self.block_size,
                  'keep_prob': self.keep_prob,
                  'sync_channels': self.sync_channels,
                  'data_format': self.data_format}
        base_config = super(DropBlock2D, self).get_config()
        return dict(list(base_config.items()) + list(config.items()))

    def compute_mask(self, inputs, mask=None):
        return mask

    def compute_output_shape(self, input_shape):
        return input_shape

    def _get_gamma(self, height, width):
        """Get the number of activation units to drop"""
        height, width = K.cast(height, K.floatx()), K.cast(width, K.floatx())
        block_size = K.constant(self.block_size, dtype=K.floatx())
        return ((1.0 - self.keep_prob) / (block_size ** 2)) *\
               (height * width / ((height - block_size + 1.0) * (width - block_size + 1.0)))

    def _compute_valid_seed_region(self, height, width):        
        positions = K.concatenate([
            tf.expand_dims(K.tile(K.expand_dims(tf.range(height), axis=1), [1, width]), axis=-1),
            tf.expand_dims(K.tile(K.expand_dims(tf.range(width), axis=0), [height, 1]), axis=-1),
        ], axis=-1)
        half_block_size = self.block_size // 2
        valid_seed_region = K.switch(
            K.all(
                K.stack(
                    [
                        positions[:, :, 0] >= half_block_size,
                        positions[:, :, 1] >= half_block_size,
                        positions[:, :, 0] < height - half_block_size,
                        positions[:, :, 1] < width - half_block_size,
                    ],
                    axis=-1,
                ),
                axis=-1,
            ),
            tf.ones((height, width)),
            tf.zeros((height, width)),
        )
        return tf.expand_dims(tf.expand_dims(valid_seed_region, axis=0), axis=-1)

    def _compute_drop_mask(self, shape):
        height, width = shape[1], shape[2]
        mask = tf.keras.backend.random_bernoulli(shape, p=self._get_gamma(height, width))
        mask *= self._compute_valid_seed_region(height, width)
        mask = tf.keras.layers.MaxPool2D(
            pool_size=(self.block_size, self.block_size),
            padding='same',
            strides=1,
            data_format='channels_last',
        )(mask)
        return 1.0 - mask

    def call(self, inputs, training=None):

        def dropped_inputs():
            outputs = inputs
            if self.data_format == 'channels_first':
                outputs = K.permute_dimensions(outputs, [0, 2, 3, 1])
            shape = K.shape(outputs)
            if self.sync_channels:
                mask = self._compute_drop_mask([shape[0], shape[1], shape[2], 1])
            else:
                mask = self._compute_drop_mask(shape)
            outputs = outputs * mask *\
                (K.cast(K.prod(shape), dtype=K.floatx()) / K.sum(mask))
            if self.data_format == 'channels_first':
                outputs = K.permute_dimensions(outputs, [0, 3, 1, 2])
            return outputs

        return K.in_train_phase(dropped_inputs, inputs, training=training)
    

def spatial_attention(input_feature):
    """
    Used in the bottleneck of SA-UNet
    """
    kernel_size = 7

    if K.image_data_format() == "channels_first":
        channel = input_feature._keras_shape[1]
        cbam_feature = Permute((2, 3, 1))(input_feature)
    else:
        # channel = input_feature._keras_shape[-1]
        channel = input_feature.shape[-1]
        cbam_feature = input_feature

    avg_pool = layers.Lambda(lambda x: K.mean(x, axis=3, keepdims=True))(cbam_feature)
    assert avg_pool.shape[-1] == 1 # assert avg_pool._keras_shape[-1] == 1
    max_pool = layers.Lambda(lambda x: K.max(x, axis=3, keepdims=True))(cbam_feature)
    assert max_pool.shape[-1] == 1 # assert max_pool._keras_shape[-1] == 1
    concat = layers.Concatenate(axis=3)([avg_pool, max_pool])
    assert max_pool.shape[-1] == 1 # assert concat._keras_shape[-1] == 2
    cbam_feature = layers.Conv2D(filters=1,
                                  kernel_size=kernel_size,
                                  strides=1,
                                  padding='same',
                                  activation='sigmoid',
                                  kernel_initializer='he_normal',
                                  use_bias=False)(concat)
    assert cbam_feature.shape[-1] == 1 # assert cbam_feature._keras_shape[-1] == 1

    if K.image_data_format() == "channels_first":
        cbam_feature = layers.Permute((3, 1, 2))(cbam_feature)

    return layers.multiply([input_feature, cbam_feature])


def conv_block(x, filter_size, size, dropout, batch_norm=False):
    """
    Convolution block of 2 Conv2D
    """
    conv = layers.Conv2D(size, (filter_size, filter_size), padding="same")(x)
    if batch_norm is True:
        conv = layers.BatchNormalization(axis=3)(conv)
    conv = layers.Activation("relu")(conv)

    conv = layers.Conv2D(size, (filter_size, filter_size), padding="same")(conv)
    if batch_norm is True:
        conv = layers.BatchNormalization(axis=3)(conv)
    conv = layers.Activation("relu")(conv)
    
    if dropout > 0:
        conv = layers.Dropout(dropout)(conv)

    return conv


def repeat_elem(tensor, rep):
    """
    lambda function to repeat Repeats the elements of a tensor along an axis
    by a factor of rep.
    If tensor has shape (None, 256,256,3), lambda will return a tensor of shape 
    (None, 256,256,6), if specified axis=3 and rep=2.
    """
    return layers.Lambda(lambda x, repnum: K.repeat_elements(x, repnum, axis=3),
                          arguments={'repnum': rep})(tensor)

    
def res_conv_block(x, filter_size, size, dropout, batch_norm=False):
    """
    Residual convolutional block.                                   
    conv - BN - Activation - conv - BN  - dropout - shortcut+BN - shortcut+BN - Activation                                     
    Check fig 4 in https://arxiv.org/ftp/arxiv/papers/1802/1802.06955.pdf
    """

    conv = layers.Conv2D(size, (filter_size, filter_size), padding='same')(x)
    if batch_norm is True:
        conv = layers.BatchNormalization(axis=3)(conv)
    conv = layers.Activation('relu')(conv)
    
    conv = layers.Conv2D(size, (filter_size, filter_size), padding='same')(conv)
    if batch_norm is True:
        conv = layers.BatchNormalization(axis=3)(conv)
    #conv = layers.Activation('relu')(conv)    #Activation before addition with shortcut
    if dropout > 0:
        conv = layers.Dropout(dropout)(conv)

    shortcut = layers.Conv2D(size, kernel_size=(1, 1), padding='same')(x)
    if batch_norm is True:
        shortcut = layers.BatchNormalization(axis=3)(shortcut)

    res_path = layers.add([shortcut, conv])
    res_path = layers.Activation('relu')(res_path)    #Activation after addition with shortcut (Original residual block)
    return res_path


def gating_signal(input, out_size, batch_norm=False):
    """
    resize the down layer feature map into the same dimension as the up layer feature map
    using 1x1 conv
    :return: the gating feature map with the same dimension of the up layer feature map
    """
    x = layers.Conv2D(out_size, (1, 1), padding='same')(input)
    if batch_norm:
        x = layers.BatchNormalization()(x)
    x = layers.Activation('relu')(x)
    return x


def attention_block(x, gating, inter_shape):
    """
    Attention gate; 
    see https://arxiv.org/pdf/1804.03999.pdf fig 2
    """
    shape_x = K.int_shape(x)
    shape_g = K.int_shape(gating)

# Getting the x signal to the same shape as the gating signal
    theta_x = layers.Conv2D(inter_shape, (2, 2), strides=(2, 2), padding='same')(x)  # 16
    shape_theta_x = K.int_shape(theta_x)

# Getting the gating signal to the same number of filters as the inter_shape
    phi_g = layers.Conv2D(inter_shape, (1, 1), padding='same')(gating)
    upsample_g = layers.Conv2DTranspose(inter_shape, (3, 3),
                                 strides=(shape_theta_x[1] // shape_g[1], shape_theta_x[2] // shape_g[2]),
                                 padding='same')(phi_g)  # 16

    concat_xg = layers.add([upsample_g, theta_x])
    act_xg = layers.Activation('relu')(concat_xg)
    psi = layers.Conv2D(1, (1, 1), padding='same')(act_xg)
    sigmoid_xg = layers.Activation('sigmoid')(psi)
    shape_sigmoid = K.int_shape(sigmoid_xg)
    upsample_psi = layers.UpSampling2D(size=(shape_x[1] // shape_sigmoid[1], shape_x[2] // shape_sigmoid[2]))(sigmoid_xg)  # 32

    upsample_psi = repeat_elem(upsample_psi, shape_x[3])

    y = layers.multiply([upsample_psi, x])

    result = layers.Conv2D(shape_x[3], (1, 1), padding='same')(y)
    result_bn = layers.BatchNormalization()(result)
    return result_bn

# models
def UNet(input_shape, NUM_CLASSES=1, dropout_rate=0.0, batch_norm=True):
    """
    UNet
    """
    # network structure
    FILTER_NUM = 64 # number of filters for the first layer
    FILTER_SIZE = 3 # size of the convolutional filter
    UP_SAMP_SIZE = 2 # size of upsampling filters
    

    inputs = layers.Input(input_shape, dtype=tf.float32)

    # Downsampling layers
    # DownRes 1, convolution + pooling
    conv_128 = conv_block(inputs, FILTER_SIZE, FILTER_NUM, dropout_rate, batch_norm)
    pool_64 = layers.MaxPooling2D(pool_size=(2,2))(conv_128)
    # DownRes 2
    conv_64 = conv_block(pool_64, FILTER_SIZE, 2*FILTER_NUM, dropout_rate, batch_norm)
    pool_32 = layers.MaxPooling2D(pool_size=(2,2))(conv_64)
    # DownRes 3
    conv_32 = conv_block(pool_32, FILTER_SIZE, 4*FILTER_NUM, dropout_rate, batch_norm)
    pool_16 = layers.MaxPooling2D(pool_size=(2,2))(conv_32)
    # DownRes 4
    conv_16 = conv_block(pool_16, FILTER_SIZE, 8*FILTER_NUM, dropout_rate, batch_norm)
    pool_8 = layers.MaxPooling2D(pool_size=(2,2))(conv_16)
    # DownRes 5, convolution only
    conv_8 = conv_block(pool_8, FILTER_SIZE, 16*FILTER_NUM, dropout_rate, batch_norm)

    # Upsampling layers
   
    up_16 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(conv_8)
    up_16 = layers.concatenate([up_16, conv_16], axis=3)
    up_conv_16 = conv_block(up_16, FILTER_SIZE, 8*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 7
    
    up_32 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_16)
    up_32 = layers.concatenate([up_32, conv_32], axis=3)
    up_conv_32 = conv_block(up_32, FILTER_SIZE, 4*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 8
    
    up_64 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_32)
    up_64 = layers.concatenate([up_64, conv_64], axis=3)
    up_conv_64 = conv_block(up_64, FILTER_SIZE, 2*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 9
   
    up_128 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_64)
    up_128 = layers.concatenate([up_128, conv_128], axis=3)
    up_conv_128 = conv_block(up_128, FILTER_SIZE, FILTER_NUM, dropout_rate, batch_norm)

    # 1*1 convolutional layers
   
    conv_final = layers.Conv2D(NUM_CLASSES, kernel_size=(1,1))(up_conv_128)
    conv_final = layers.BatchNormalization(axis=3)(conv_final)
    conv_final = layers.Activation('sigmoid')(conv_final)  #Change to softmax for multichannel

    # Model 
    model = models.Model(inputs, conv_final, name="UNet")
    #print(model.summary())
    return model


def Attention_UNet(input_shape, NUM_CLASSES=1, dropout_rate=0.0, batch_norm=True):
    """
    Attention UNet
    """
    # network structure
    FILTER_NUM = 64 # number of basic filters for the first layer
    FILTER_SIZE = 3 # size of the convolutional filter
    UP_SAMP_SIZE = 2 # size of upsampling filters
    
    inputs = layers.Input(input_shape, dtype=tf.float32)

    # Downsampling layers
    # DownRes 1, convolution + pooling
    conv_128 = conv_block(inputs, FILTER_SIZE, FILTER_NUM, dropout_rate, batch_norm)
    pool_64 = layers.MaxPooling2D(pool_size=(2,2))(conv_128)
    # DownRes 2
    conv_64 = conv_block(pool_64, FILTER_SIZE, 2*FILTER_NUM, dropout_rate, batch_norm)
    pool_32 = layers.MaxPooling2D(pool_size=(2,2))(conv_64)
    # DownRes 3
    conv_32 = conv_block(pool_32, FILTER_SIZE, 4*FILTER_NUM, dropout_rate, batch_norm)
    pool_16 = layers.MaxPooling2D(pool_size=(2,2))(conv_32)
    # DownRes 4
    conv_16 = conv_block(pool_16, FILTER_SIZE, 8*FILTER_NUM, dropout_rate, batch_norm)
    pool_8 = layers.MaxPooling2D(pool_size=(2,2))(conv_16)
    # DownRes 5, convolution only
    conv_8 = conv_block(pool_8, FILTER_SIZE, 16*FILTER_NUM, dropout_rate, batch_norm)

    # Upsampling layers
    # UpRes 6, attention gated concatenation + upsampling + double residual convolution
    gating_16 = gating_signal(conv_8, 8*FILTER_NUM, batch_norm)
    att_16 = attention_block(conv_16, gating_16, 8*FILTER_NUM)
    up_16 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(conv_8)
    up_16 = layers.concatenate([up_16, att_16], axis=3)
    up_conv_16 = conv_block(up_16, FILTER_SIZE, 8*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 7
    gating_32 = gating_signal(up_conv_16, 4*FILTER_NUM, batch_norm)
    att_32 = attention_block(conv_32, gating_32, 4*FILTER_NUM)
    up_32 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_16)
    up_32 = layers.concatenate([up_32, att_32], axis=3)
    up_conv_32 = conv_block(up_32, FILTER_SIZE, 4*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 8
    gating_64 = gating_signal(up_conv_32, 2*FILTER_NUM, batch_norm)
    att_64 = attention_block(conv_64, gating_64, 2*FILTER_NUM)
    up_64 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_32)
    up_64 = layers.concatenate([up_64, att_64], axis=3)
    up_conv_64 = conv_block(up_64, FILTER_SIZE, 2*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 9
    gating_128 = gating_signal(up_conv_64, FILTER_NUM, batch_norm)
    att_128 = attention_block(conv_128, gating_128, FILTER_NUM)
    up_128 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_64)
    up_128 = layers.concatenate([up_128, att_128], axis=3)
    up_conv_128 = conv_block(up_128, FILTER_SIZE, FILTER_NUM, dropout_rate, batch_norm)

    # 1*1 convolutional layers
    conv_final = layers.Conv2D(NUM_CLASSES, kernel_size=(1,1))(up_conv_128)
    conv_final = layers.BatchNormalization(axis=3)(conv_final)
    conv_final = layers.Activation('sigmoid')(conv_final)  #Change to softmax for multichannel

    # Model integration
    model = models.Model(inputs, conv_final, name="Attention_UNet")
    return model


def Attention_ResUNet(input_shape, NUM_CLASSES=1, dropout_rate=0.0, batch_norm=True):
    """
    Attention Residual UNet
    """
    # network structure
    FILTER_NUM = 64 # number of basic filters for the first layer
    FILTER_SIZE = 3 # size of the convolutional filter
    UP_SAMP_SIZE = 2 # size of upsampling filters
    # input data
    # dimension of the image depth
    inputs = layers.Input(input_shape, dtype=tf.float32)
    axis = 3

    # Downsampling layers
    # DownRes 1, double residual convolution + pooling
    conv_128 = res_conv_block(inputs, FILTER_SIZE, FILTER_NUM, dropout_rate, batch_norm)
    pool_64 = layers.MaxPooling2D(pool_size=(2,2))(conv_128)
    # DownRes 2
    conv_64 = res_conv_block(pool_64, FILTER_SIZE, 2*FILTER_NUM, dropout_rate, batch_norm)
    pool_32 = layers.MaxPooling2D(pool_size=(2,2))(conv_64)
    # DownRes 3
    conv_32 = res_conv_block(pool_32, FILTER_SIZE, 4*FILTER_NUM, dropout_rate, batch_norm)
    pool_16 = layers.MaxPooling2D(pool_size=(2,2))(conv_32)
    # DownRes 4
    conv_16 = res_conv_block(pool_16, FILTER_SIZE, 8*FILTER_NUM, dropout_rate, batch_norm)
    pool_8 = layers.MaxPooling2D(pool_size=(2,2))(conv_16)
    # DownRes 5, convolution only
    conv_8 = res_conv_block(pool_8, FILTER_SIZE, 16*FILTER_NUM, dropout_rate, batch_norm)

    # Upsampling layers
    # UpRes 6, attention gated concatenation + upsampling + double residual convolution
    gating_16 = gating_signal(conv_8, 8*FILTER_NUM, batch_norm)
    att_16 = attention_block(conv_16, gating_16, 8*FILTER_NUM)
    up_16 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(conv_8)
    up_16 = layers.concatenate([up_16, att_16], axis=axis)
    up_conv_16 = res_conv_block(up_16, FILTER_SIZE, 8*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 7
    gating_32 = gating_signal(up_conv_16, 4*FILTER_NUM, batch_norm)
    att_32 = attention_block(conv_32, gating_32, 4*FILTER_NUM)
    up_32 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_16)
    up_32 = layers.concatenate([up_32, att_32], axis=axis)
    up_conv_32 = res_conv_block(up_32, FILTER_SIZE, 4*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 8
    gating_64 = gating_signal(up_conv_32, 2*FILTER_NUM, batch_norm)
    att_64 = attention_block(conv_64, gating_64, 2*FILTER_NUM)
    up_64 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_32)
    up_64 = layers.concatenate([up_64, att_64], axis=axis)
    up_conv_64 = res_conv_block(up_64, FILTER_SIZE, 2*FILTER_NUM, dropout_rate, batch_norm)
    # UpRes 9
    gating_128 = gating_signal(up_conv_64, FILTER_NUM, batch_norm)
    att_128 = attention_block(conv_128, gating_128, FILTER_NUM)
    up_128 = layers.UpSampling2D(size=(UP_SAMP_SIZE, UP_SAMP_SIZE), data_format="channels_last")(up_conv_64)
    up_128 = layers.concatenate([up_128, att_128], axis=axis)
    up_conv_128 = res_conv_block(up_128, FILTER_SIZE, FILTER_NUM, dropout_rate, batch_norm)

    # 1*1 convolutional layers
    
    conv_final = layers.Conv2D(NUM_CLASSES, kernel_size=(1,1))(up_conv_128)
    conv_final = layers.BatchNormalization(axis=axis)(conv_final)
    conv_final = layers.Activation('sigmoid')(conv_final)  #Change to softmax for multichannel

    # Model integration
    model = models.Model(inputs, conv_final, name="AttentionResUNet")
    return model


def SA_UNet(input_shape, block_size=7, keep_prob=0.9, start_neurons=16):
    """
    Spatial Attention U-Net
    """
    inputs = layers.Input(input_shape)
    conv1 = layers.Conv2D(start_neurons * 1, (3, 3), activation=None, padding="same")(inputs)
    conv1 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(conv1)
    conv1= layers.BatchNormalization()(conv1)
    conv1 = layers.Activation('relu')(conv1)
    conv1 = layers.Conv2D(start_neurons * 1, (3, 3), activation=None, padding="same")(conv1)
    conv1 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(conv1)
    conv1 = layers.BatchNormalization()(conv1)
    conv1 = layers.Activation('relu')(conv1)
    pool1 = layers.MaxPooling2D((2, 2))(conv1)

    conv2 = layers.Conv2D(start_neurons * 2, (3, 3), activation=None, padding="same")(pool1)
    conv2 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(conv2)
    conv2 = layers.BatchNormalization()(conv2)
    conv2 = layers.Activation('relu')(conv2)

    conv2 = layers.Conv2D(start_neurons * 2, (3, 3), activation=None, padding="same")(conv2)
    conv2 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(conv2)
    conv2 = layers.BatchNormalization()(conv2)
    conv2 = layers.Activation('relu')(conv2)
    pool2 = layers.MaxPooling2D((2, 2))(conv2)


    conv3 = layers.Conv2D(start_neurons * 4, (3, 3), activation=None, padding="same")(pool2)
    conv3 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(conv3)
    conv3 = layers.BatchNormalization()(conv3)
    conv3 = layers.Activation('relu')(conv3)
    conv3 = layers.Conv2D(start_neurons * 4, (3, 3), activation=None, padding="same")(conv3)
    conv3 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(conv3)
    conv3 = layers.BatchNormalization()(conv3)
    conv3 = layers.Activation('relu')(conv3)
    pool3 = layers.MaxPooling2D((2, 2))(conv3)


    convm = layers.Conv2D(start_neurons * 8, (3, 3), activation=None, padding="same")(pool3)
    convm = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(convm)
    convm = layers.BatchNormalization()(convm)
    convm = layers.Activation('relu')(convm)
    convm = spatial_attention(convm)
    convm = layers.Conv2D(start_neurons * 8, (3, 3), activation=None, padding="same")(convm)
    convm = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(convm)
    convm = layers.BatchNormalization()(convm)
    convm = layers.Activation('relu')(convm)


    deconv3 = layers.Conv2DTranspose(start_neurons * 4, (3, 3), strides=(2, 2), padding="same")(convm)
    uconv3 = layers.concatenate([deconv3, conv3])

    uconv3 = layers.Conv2D(start_neurons * 4, (3, 3), activation=None, padding="same")(uconv3)
    uconv3 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(uconv3)
    uconv3 = layers.BatchNormalization()(uconv3)
    uconv3 = layers.Activation('relu')(uconv3)
    uconv3 = layers.Conv2D(start_neurons * 4, (3, 3), activation=None, padding="same")(uconv3)
    uconv3 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(uconv3)
    uconv3 = layers.BatchNormalization()(uconv3)
    uconv3 = layers.Activation('relu')(uconv3)

    deconv2 = layers.Conv2DTranspose(start_neurons * 2, (3, 3), strides=(2, 2), padding="same")(uconv3)
    uconv2 = layers.concatenate([deconv2, conv2])

    uconv2 = layers.Conv2D(start_neurons * 2, (3, 3), activation=None, padding="same")(uconv2)
    uconv2 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(uconv2)
    uconv2 = layers.BatchNormalization()(uconv2)
    uconv2 = layers.Activation('relu')(uconv2)
    uconv2 = layers.Conv2D(start_neurons * 2, (3, 3), activation=None, padding="same")(uconv2)
    uconv2 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(uconv2)
    uconv2 = layers.BatchNormalization()(uconv2)
    uconv2 = layers.Activation('relu')(uconv2)

    deconv1 = layers.Conv2DTranspose(start_neurons * 1, (3, 3), strides=(2, 2), padding="same")(uconv2)
    uconv1 = layers.concatenate([deconv1, conv1])


    uconv1 = layers.Conv2D(start_neurons * 1, (3, 3), activation=None, padding="same")(uconv1)
    uconv1 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(uconv1)
    uconv1 = layers.BatchNormalization()(uconv1)
    uconv1 = layers.Activation('relu')(uconv1)
    uconv1 = layers.Conv2D(start_neurons * 1, (3, 3), activation=None, padding="same")(uconv1)
    uconv1 = DropBlock2D(block_size=block_size, keep_prob=keep_prob)(uconv1)
    uconv1 = layers.BatchNormalization()(uconv1)
    uconv1 = layers.Activation('relu')(uconv1)
    output_layer_noActi = layers.Conv2D(1, (1, 1), padding="same", activation=None)(uconv1)
    output_layer = layers.Activation('sigmoid')(output_layer_noActi)

    model = models.Model(inputs=inputs, outputs=output_layer, name="SA_UNet")
    
    #     model.compile(optimizer=Adam(learning_rate=1e-3),
    #                   loss='binary_crossentropy',
    #                   metrics=['accuracy',[dice_coef, mse_euclidean])

    return model

# fuctions for model analysis
def plot_history(history, title=""):
    """
    Plots a trainning hsitory
    """
    fig = plt.figure(figsize=(16, 8))
    fig.suptitle(title, fontsize=16)
    
    plt.subplot(1, 3, 1)
    loss = history.history['loss']
    val_loss = history.history['val_loss']
    epochs = range(1, len(loss) + 1)
    plt.plot(epochs, loss, 'y', label='Training loss')
    plt.plot(epochs, val_loss, 'r', label='Validation loss')
    plt.title('Training and validation loss')
    plt.xlabel('Epochs')
    plt.ylabel('Loss')
    plt.ylim(0, 0.5)
    plt.legend()
    plt.grid()

    plt.subplot(1, 3, 2)    
    plt.plot(epochs, history.history['mae_euclidean'], 'y', label='mae_euclidean')
    plt.plot(epochs, history.history['val_mae_euclidean'], 'r', label='val_mae_euclidean')
    plt.title('MAE euclidean distances from borders')
    plt.xlabel('Epochs')
    plt.ylabel('MAE euclidean')
    plt.ylim(0, 10)
    plt.legend()
    plt.grid()
        
    plt.subplot(1, 3, 3)
    plt.plot(epochs, history.history['jacard_coef'], 'y', label='jacard_coef')
    plt.plot(epochs, history.history['val_jacard_coef'], 'r', label='val_jacard_coef')
    plt.title('Training and validation jacard coef')
    plt.xlabel('Epochs')
    plt.ylabel('jacard coef')
    plt.ylim(0, 0.8)
    plt.legend()
    plt.grid()
        
    plt.show()
    
    
def visual_eval(model, test_img, mask_expected=None, thresh = 0.5, show=True):
    """
    Provides visual evaluation for the performance of a model.
    """
    test_img = img_as_float32(test_img).copy()
    original_img = test_img[:, :, 0].copy()

    X = np.expand_dims(test_img, axis=0)
    #     X = np.expand_dims(X, axis=-1)
    mask_predicted = model.predict(X)[0, :, :, 0]
    mask_predicted = mask_predicted > thresh # thresholding
    mask_expected = mask_expected[:, :, 0]
    
    img_review = color.gray2rgb(original_img)
    if mask_expected is not None:
        mask_expected = mask_expected.astype(bool)
        img_review[mask_expected & mask_predicted] = (0, 1, 0) # tp, green
        img_review[mask_expected & ~mask_predicted] = (0, 0, 1) # fn, low recall score
        img_review[~mask_expected & mask_predicted] = (1, 1, 0) # fp, low precision score
    else:
        img_review[mask_predicted] = (0, 1, 0) # predicted, 
    
    # adding legend
    cv2.putText(img_review, "TP", (5, 10), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 1, 0), 1)
    cv2.putText(img_review, "FP", (18, 10), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (1, 1, 0), 1)
    cv2.putText(img_review, "FN", (31, 10), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 1), 1)
    
    if show:
        plt.figure(figsize=(12,12))
        plt.imshow(img_review)
        plt.title("Mark-up picture")
        plt.show()

    return img_review #, mask_expected, mask_predicted


def plot_train_val(model, idx_train=None, idx_val=None):
    # plot prediction accuracy on train and validation dataset
    if idx_train == None:
        idx_train = np.random.randint(0, X_train.shape[0])
    if idx_val == None:
        idx_val = np.random.randint(0, X_val.shape[0])
    
    plt.figure(figsize=(18,10))

    plt.subplot(1, 2, 1)
    im = visual_eval(model=model, test_img=X_train[idx_train], mask_expected=y_train[idx_train], show=False)
    plt.imshow(im)
    plt.title(f"Train image {idx_train}")

    plt.subplot(1, 2, 2)
    im = visual_eval(model=model, test_img=X_val[idx_val], mask_expected=y_val[idx_val], show=False)
    plt.imshow(im)
    plt.title(f"Validation image {idx_val}")

    plt.show()

Train and test data loading

Train and test datasets are organized in separate folders. The dataset is prepared from one image with scale 500x and a second image 1000x. The images are sliced into patches with 256x176. Augmentation is applied to all patches by applying median filter. The patches are divided so that patch and its augmentation are either in train, validation or test; this is done to avoid information leak. Augmentation could be automated with data generator, but it is possible some augmentation settings to produce unrealistic images. It is strongly recommended to check the visual perception of the augmentation if such approach is considered.

def create_augmneted_img(path):
    """
    Function to create augmented images by appling median filter
    """
    for f_name in os.listdir(path):
        if f_name.split("_")[-1] == 'image.tif' and f_name.startswith("patch1000_"):
            print(f"Working on {f_name}")
            img = io.imread(train_test_path + r"\\" + f_name)
            img = img_as_float32(img)
            aug = filters.median(img)
            new_name = str(train_test_path + r"\\" + f_name).replace("patch1000_", "patch1000_aug")
            io.imsave(new_name, img_as_ubyte(aug))

            mask_name = f_name.replace("image", "mask")
            mask = io.imread(train_test_path + r"\\" + mask_name)
            new_name = new_name.replace("_image", "_mask")
            io.imsave(new_name, img_as_ubyte(mask))
            
#create_augmneted_img(r"datasets/train") # already done

def load_data(path):
    """
    loads images and masks from a folder
    all images and mask should be 256x176 (w x h)
    """
    imgs = []
    masks = []
    for f_name in os.listdir(path):
        if f_name.split("_")[-1] == 'image.tif':
            # print(f"Working on {f_name}")
            img = io.imread(path + r"\\" + f_name)
            assert img.shape == (176, 256)
            img = img_as_float32(img)
            imgs.append(img)

            mask_name = f_name.replace("image", "mask")
            mask = io.imread(path + r"\\" + mask_name)
            mask = mask // 255
            masks.append(mask)
    imgs = np.expand_dims(np.array(imgs), -1)
    masks = np.expand_dims(np.array(masks), -1)
    return imgs, masks

X_train, y_train = load_data("datasets/train")
X_val, y_val = load_data("datasets/validation")
X_test, y_test = load_data("datasets/test")

X_train = tf.cast(X_train, tf.float16)
X_val = tf.cast(X_val, tf.float16)
X_test = tf.cast(X_test, tf.float16)

X_train.shape, X_val.shape,  X_test.shape

(TensorShape([32, 176, 256, 1]),
 TensorShape([8, 176, 256, 1]),
 TensorShape([4, 176, 256, 1]))

# sanity check for mating of mages and masks; run a couple of times
# t = np.random.randint(X_train.shape[0])
t = np.random.randint(X_train.shape[0])
plt.figure(figsize=(12,5))
plt.subplot(1, 2, 1)
plt.imshow(tf.cast(X_train[t], float), cmap='gray')
plt.subplot(1, 2, 2)
plt.imshow(y_train[t], cmap='gray')
plt.show()

total_border_pixels = np.sum(y_train)
total_pixels = y_train.shape[0] * y_train.shape[1] * y_train.shape[2] * y_train.shape[3]
border_pix_ratio = total_border_pixels / total_pixels
print(f"Grain border pixels / all pixels = {border_pix_ratio}")

Grain border pixels / all pixels = 0.07478609952059659

IMG_HEIGHT = X_train.shape[1]
IMG_WIDTH  = X_train.shape[2]
IMG_CHANNELS = X_train.shape[3]
input_shape = (IMG_HEIGHT,IMG_WIDTH,IMG_CHANNELS)
input_shape

(176, 256, 1)

Model selection

Several iterations and hyperparameter searches have been done in the background. The next sections will start with setting known to produce good results.
Objectives:

• maximize Jacard coefficient; Jacard coefficient is presenting IoU
• minimize the MAE euclidean distances; Euclidean distances are calculated for the ground truth and the predicted border mask. The differences between the distances is measured by MAE. It is preferred to have low and stable MAE euclidean distances. MAE euclidean distance error is not stable by nature, because even one FP pixel in the middle of a grain could increase the metric significantly. MAE euclidean distances are used in the postprocessing step by watershed algorithm, making the metric important.
• prevent overfitting

SA UNet model trials

As a starting point, the following parameters have been found to be optimal:block_size=19, keep_prob=0.8, start_neurons=20

# SA Unet1
tf.random.set_seed(14)
model_sa_unet1 = SA_UNet(input_shape, block_size=19, keep_prob=0.8, start_neurons=20)
model_sa_unet1.compile(optimizer=Adam(learning_rate=1e-2),
                         loss = dice_coef_loss,
                         metrics=[mae_euclidean, jacard_coef]
                         )
start = time.time()
history_sa_unet1 = model_sa_unet1.fit(X_train, y_train,
                                    validation_data = (X_val, y_val),
                                    batch_size = 8, # no resources for 16
                                    epochs=200
                                   )
print(f"Model trained for {time.time() - start}s")
model_sa_unet1.save("2022-02-17 SA-UNet1 200epochs.hdf5")

Epoch 1/200
4/4 [==============================] - 10s 479ms/step - loss: 0.8405 - mae_euclidean: 5.2724 - jacard_coef: 0.0874 - val_loss: 0.8714 - val_mae_euclidean: 5.8165 - val_jacard_coef: 0.0687
Epoch 2/200
4/4 [==============================] - 1s 374ms/step - loss: 0.7713 - mae_euclidean: 3.6544 - jacard_coef: 0.1293 - val_loss: 0.8798 - val_mae_euclidean: 5.6584 - val_jacard_coef: 0.0640
Epoch 3/200
4/4 [==============================] - 1s 376ms/step - loss: 0.7341 - mae_euclidean: 3.3476 - jacard_coef: 0.1534 - val_loss: 0.9941 - val_mae_euclidean: 121.6934 - val_jacard_coef: 0.0030
Epoch 4/200
4/4 [==============================] - 1s 373ms/step - loss: 0.6905 - mae_euclidean: 2.9527 - jacard_coef: 0.1831 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.6859e-05
Epoch 5/200
4/4 [==============================] - 1s 374ms/step - loss: 0.6529 - mae_euclidean: 2.8639 - jacard_coef: 0.2107 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.6988e-05
Epoch 6/200
4/4 [==============================] - 1s 372ms/step - loss: 0.6078 - mae_euclidean: 2.5158 - jacard_coef: 0.2441 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.2680e-05
Epoch 7/200
4/4 [==============================] - 1s 376ms/step - loss: 0.5741 - mae_euclidean: 2.2905 - jacard_coef: 0.2715 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.8226e-05
Epoch 8/200
4/4 [==============================] - 1s 376ms/step - loss: 0.5310 - mae_euclidean: 1.9890 - jacard_coef: 0.3067 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.6139e-05
Epoch 9/200
4/4 [==============================] - 1s 377ms/step - loss: 0.4989 - mae_euclidean: 1.8818 - jacard_coef: 0.3346 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.8324e-05
Epoch 10/200
4/4 [==============================] - 1s 378ms/step - loss: 0.4689 - mae_euclidean: 1.5774 - jacard_coef: 0.3616 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.7041e-05
Epoch 11/200
4/4 [==============================] - 1s 373ms/step - loss: 0.4546 - mae_euclidean: 1.3611 - jacard_coef: 0.3754 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.7742e-05
Epoch 12/200
4/4 [==============================] - 1s 376ms/step - loss: 0.4356 - mae_euclidean: 1.6012 - jacard_coef: 0.3932 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.6924e-05
Epoch 13/200
4/4 [==============================] - 1s 375ms/step - loss: 0.4289 - mae_euclidean: 1.5228 - jacard_coef: 0.3997 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.6965e-05
Epoch 14/200
4/4 [==============================] - 1s 376ms/step - loss: 0.4259 - mae_euclidean: 1.5623 - jacard_coef: 0.4027 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.6926e-05
Epoch 15/200
4/4 [==============================] - 1s 374ms/step - loss: 0.4099 - mae_euclidean: 1.5322 - jacard_coef: 0.4186 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.8929e-05
Epoch 16/200
4/4 [==============================] - 1s 375ms/step - loss: 0.4056 - mae_euclidean: 1.4564 - jacard_coef: 0.4229 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.1543e-05
Epoch 17/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3996 - mae_euclidean: 1.4724 - jacard_coef: 0.4290 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.9352e-05
Epoch 18/200
4/4 [==============================] - 1s 374ms/step - loss: 0.4029 - mae_euclidean: 1.5331 - jacard_coef: 0.4258 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.2381e-04
Epoch 19/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3967 - mae_euclidean: 1.5450 - jacard_coef: 0.4320 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.6911e-05
Epoch 20/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3903 - mae_euclidean: 1.3981 - jacard_coef: 0.4388 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.1258e-04
Epoch 21/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3815 - mae_euclidean: 1.2309 - jacard_coef: 0.4479 - val_loss: 0.9896 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 0.0053
Epoch 22/200
4/4 [==============================] - 1s 376ms/step - loss: 0.3747 - mae_euclidean: 1.4530 - jacard_coef: 0.4550 - val_loss: 0.9994 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.3946e-04
Epoch 23/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3825 - mae_euclidean: 1.4358 - jacard_coef: 0.4473 - val_loss: 0.9995 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.5189e-04
Epoch 24/200
4/4 [==============================] - 1s 376ms/step - loss: 0.3696 - mae_euclidean: 1.3172 - jacard_coef: 0.4604 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.4790e-05
Epoch 25/200
4/4 [==============================] - 1s 374ms/step - loss: 0.3672 - mae_euclidean: 1.2925 - jacard_coef: 0.4629 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.8606e-05
Epoch 26/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3596 - mae_euclidean: 1.1674 - jacard_coef: 0.4711 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.2000e-05
Epoch 27/200
4/4 [==============================] - 1s 372ms/step - loss: 0.3683 - mae_euclidean: 1.3403 - jacard_coef: 0.4622 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.9069e-05
Epoch 28/200
4/4 [==============================] - 1s 374ms/step - loss: 0.3690 - mae_euclidean: 1.6199 - jacard_coef: 0.4620 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.5954e-05
Epoch 29/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3567 - mae_euclidean: 1.1713 - jacard_coef: 0.4742 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.8024e-05
Epoch 30/200
4/4 [==============================] - 1s 376ms/step - loss: 0.3573 - mae_euclidean: 1.2792 - jacard_coef: 0.4738 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.4576e-05
Epoch 31/200
4/4 [==============================] - 1s 378ms/step - loss: 0.3467 - mae_euclidean: 1.0636 - jacard_coef: 0.4853 - val_loss: 0.9995 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.5842e-04
Epoch 32/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3529 - mae_euclidean: 1.2393 - jacard_coef: 0.4784 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.9170e-05
Epoch 33/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3549 - mae_euclidean: 1.2285 - jacard_coef: 0.4763 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.6148e-05
Epoch 34/200
4/4 [==============================] - 1s 378ms/step - loss: 0.3449 - mae_euclidean: 1.1468 - jacard_coef: 0.4877 - val_loss: 0.9996 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.3426e-04
Epoch 35/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3549 - mae_euclidean: 1.4855 - jacard_coef: 0.4774 - val_loss: 0.9996 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.9960e-04
Epoch 36/200
4/4 [==============================] - 1s 374ms/step - loss: 0.3306 - mae_euclidean: 1.0508 - jacard_coef: 0.5031 - val_loss: 0.9997 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.8476e-04
Epoch 37/200
4/4 [==============================] - 1s 378ms/step - loss: 0.3297 - mae_euclidean: 1.1267 - jacard_coef: 0.5041 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.3156e-05
Epoch 38/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3298 - mae_euclidean: 1.0918 - jacard_coef: 0.5042 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.3458e-05
Epoch 39/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3234 - mae_euclidean: 1.1033 - jacard_coef: 0.5114 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.5992e-05
Epoch 40/200
4/4 [==============================] - 1s 376ms/step - loss: 0.3261 - mae_euclidean: 1.2548 - jacard_coef: 0.5083 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.0925e-04
Epoch 41/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3270 - mae_euclidean: 1.1142 - jacard_coef: 0.5075 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.9689e-05
Epoch 42/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3267 - mae_euclidean: 1.2009 - jacard_coef: 0.5076 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.2712e-04
Epoch 43/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3183 - mae_euclidean: 1.0474 - jacard_coef: 0.5173 - val_loss: 0.9997 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.6183e-04
Epoch 44/200
4/4 [==============================] - 1s 380ms/step - loss: 0.3276 - mae_euclidean: 1.1611 - jacard_coef: 0.5068 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.6311e-05
Epoch 45/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3270 - mae_euclidean: 1.1466 - jacard_coef: 0.5074 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0383e-05
Epoch 46/200
4/4 [==============================] - 1s 378ms/step - loss: 0.3403 - mae_euclidean: 1.1252 - jacard_coef: 0.4926 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.4382e-05
Epoch 47/200
4/4 [==============================] - 1s 376ms/step - loss: 0.3239 - mae_euclidean: 1.0483 - jacard_coef: 0.5109 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.5168e-05
Epoch 48/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3180 - mae_euclidean: 1.1810 - jacard_coef: 0.5178 - val_loss: 0.9990 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0349e-04
Epoch 49/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3216 - mae_euclidean: 1.0358 - jacard_coef: 0.5135 - val_loss: 0.9990 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.9472e-04
Epoch 50/200
4/4 [==============================] - 1s 376ms/step - loss: 0.3167 - mae_euclidean: 1.1031 - jacard_coef: 0.5191 - val_loss: 0.9994 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.4288e-04
Epoch 51/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3159 - mae_euclidean: 1.0458 - jacard_coef: 0.5200 - val_loss: 0.9996 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.1818e-04
Epoch 52/200
4/4 [==============================] - 1s 378ms/step - loss: 0.3111 - mae_euclidean: 0.9833 - jacard_coef: 0.5254 - val_loss: 0.9996 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.2130e-04
Epoch 53/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3123 - mae_euclidean: 1.0192 - jacard_coef: 0.5240 - val_loss: 0.9996 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.9369e-04
Epoch 54/200
4/4 [==============================] - 1s 374ms/step - loss: 0.3082 - mae_euclidean: 1.0124 - jacard_coef: 0.5289 - val_loss: 0.9997 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.8522e-04
Epoch 55/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3057 - mae_euclidean: 0.9631 - jacard_coef: 0.5318 - val_loss: 0.9995 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.8492e-04
Epoch 56/200
4/4 [==============================] - 1s 378ms/step - loss: 0.3085 - mae_euclidean: 1.0028 - jacard_coef: 0.5285 - val_loss: 0.9996 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.0432e-04
Epoch 57/200
4/4 [==============================] - 1s 374ms/step - loss: 0.3068 - mae_euclidean: 1.0010 - jacard_coef: 0.5306 - val_loss: 0.9997 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.7910e-04
Epoch 58/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3041 - mae_euclidean: 1.0141 - jacard_coef: 0.5338 - val_loss: 0.9990 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.2869e-04
Epoch 59/200
4/4 [==============================] - 1s 374ms/step - loss: 0.3069 - mae_euclidean: 1.0177 - jacard_coef: 0.5304 - val_loss: 0.9996 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.1351e-04
Epoch 60/200
4/4 [==============================] - 1s 374ms/step - loss: 0.3173 - mae_euclidean: 1.0753 - jacard_coef: 0.5186 - val_loss: 0.9994 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.4152e-04
Epoch 61/200
4/4 [==============================] - 1s 378ms/step - loss: 0.3019 - mae_euclidean: 1.0748 - jacard_coef: 0.5363 - val_loss: 0.9993 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 3.7767e-04
Epoch 62/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3090 - mae_euclidean: 1.0858 - jacard_coef: 0.5281 - val_loss: 0.9990 - val_mae_euclidean: 13835056955770535936.0000 - val_jacard_coef: 5.2017e-04
Epoch 63/200
4/4 [==============================] - 1s 380ms/step - loss: 0.3164 - mae_euclidean: 1.1287 - jacard_coef: 0.5193 - val_loss: 0.9994 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 3.3474e-04
Epoch 64/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3293 - mae_euclidean: 1.0888 - jacard_coef: 0.5055 - val_loss: 0.9987 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 6.7962e-04
Epoch 65/200
4/4 [==============================] - 1s 377ms/step - loss: 0.3135 - mae_euclidean: 1.0868 - jacard_coef: 0.5227 - val_loss: 0.9989 - val_mae_euclidean: 6917528477885267968.0000 - val_jacard_coef: 5.9354e-04
Epoch 66/200
4/4 [==============================] - 1s 376ms/step - loss: 0.3235 - mae_euclidean: 1.1420 - jacard_coef: 0.5113 - val_loss: 0.9991 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 4.9040e-04
Epoch 67/200
4/4 [==============================] - 1s 375ms/step - loss: 0.3116 - mae_euclidean: 1.1812 - jacard_coef: 0.5249 - val_loss: 0.9967 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0017
Epoch 68/200
4/4 [==============================] - 1s 372ms/step - loss: 0.3095 - mae_euclidean: 1.0380 - jacard_coef: 0.5273 - val_loss: 0.9940 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0030
Epoch 69/200
4/4 [==============================] - 1s 373ms/step - loss: 0.3020 - mae_euclidean: 1.1099 - jacard_coef: 0.5365 - val_loss: 0.9979 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0011
Epoch 70/200
4/4 [==============================] - 1s 378ms/step - loss: 0.2963 - mae_euclidean: 0.9936 - jacard_coef: 0.5429 - val_loss: 0.9987 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 6.5136e-04
Epoch 71/200
4/4 [==============================] - 2s 394ms/step - loss: 0.3028 - mae_euclidean: 1.2029 - jacard_coef: 0.5356 - val_loss: 0.9980 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 0.0010
Epoch 72/200
4/4 [==============================] - 2s 401ms/step - loss: 0.3048 - mae_euclidean: 1.1255 - jacard_coef: 0.5331 - val_loss: 0.9986 - val_mae_euclidean: 6917528477885267968.0000 - val_jacard_coef: 7.2250e-04
Epoch 73/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2945 - mae_euclidean: 1.0284 - jacard_coef: 0.5450 - val_loss: 0.9986 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 7.2496e-04
Epoch 74/200
4/4 [==============================] - 2s 396ms/step - loss: 0.3086 - mae_euclidean: 1.0777 - jacard_coef: 0.5285 - val_loss: 0.9986 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 7.0706e-04
Epoch 75/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2924 - mae_euclidean: 1.0098 - jacard_coef: 0.5476 - val_loss: 0.9874 - val_mae_euclidean: 32.9031 - val_jacard_coef: 0.0064
Epoch 76/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2940 - mae_euclidean: 1.0228 - jacard_coef: 0.5458 - val_loss: 0.9951 - val_mae_euclidean: 56.3397 - val_jacard_coef: 0.0025
Epoch 77/200
4/4 [==============================] - 2s 397ms/step - loss: 0.3053 - mae_euclidean: 1.3810 - jacard_coef: 0.5331 - val_loss: 0.9901 - val_mae_euclidean: 47.0294 - val_jacard_coef: 0.0050
Epoch 78/200
4/4 [==============================] - 2s 396ms/step - loss: 0.3022 - mae_euclidean: 1.2327 - jacard_coef: 0.5359 - val_loss: 0.9946 - val_mae_euclidean: 56.1145 - val_jacard_coef: 0.0027
Epoch 79/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2932 - mae_euclidean: 0.9684 - jacard_coef: 0.5468 - val_loss: 0.9956 - val_mae_euclidean: 77.2754 - val_jacard_coef: 0.0022
Epoch 80/200
4/4 [==============================] - 2s 399ms/step - loss: 0.3009 - mae_euclidean: 1.1001 - jacard_coef: 0.5379 - val_loss: 0.9836 - val_mae_euclidean: 35.1385 - val_jacard_coef: 0.0083
Epoch 81/200
4/4 [==============================] - 2s 400ms/step - loss: 0.2968 - mae_euclidean: 0.9880 - jacard_coef: 0.5424 - val_loss: 0.9612 - val_mae_euclidean: 23.4549 - val_jacard_coef: 0.0198
Epoch 82/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2856 - mae_euclidean: 1.0051 - jacard_coef: 0.5558 - val_loss: 0.9851 - val_mae_euclidean: 36.7373 - val_jacard_coef: 0.0075
Epoch 83/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2850 - mae_euclidean: 1.0334 - jacard_coef: 0.5565 - val_loss: 0.9665 - val_mae_euclidean: 25.8636 - val_jacard_coef: 0.0170
Epoch 84/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2883 - mae_euclidean: 0.9375 - jacard_coef: 0.5527 - val_loss: 0.9746 - val_mae_euclidean: 26.3080 - val_jacard_coef: 0.0129
Epoch 85/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2908 - mae_euclidean: 0.9924 - jacard_coef: 0.5495 - val_loss: 0.9658 - val_mae_euclidean: 20.6509 - val_jacard_coef: 0.0174
Epoch 86/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2890 - mae_euclidean: 1.1424 - jacard_coef: 0.5517 - val_loss: 0.9250 - val_mae_euclidean: 18.0836 - val_jacard_coef: 0.0390
Epoch 87/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2856 - mae_euclidean: 0.9364 - jacard_coef: 0.5559 - val_loss: 0.9091 - val_mae_euclidean: 13.3445 - val_jacard_coef: 0.0476
Epoch 88/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2819 - mae_euclidean: 1.0285 - jacard_coef: 0.5603 - val_loss: 0.8696 - val_mae_euclidean: 11.2002 - val_jacard_coef: 0.0698
Epoch 89/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2924 - mae_euclidean: 1.2375 - jacard_coef: 0.5481 - val_loss: 0.8157 - val_mae_euclidean: 9.3213 - val_jacard_coef: 0.1015
Epoch 90/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2864 - mae_euclidean: 1.0620 - jacard_coef: 0.5548 - val_loss: 0.8483 - val_mae_euclidean: 9.8455 - val_jacard_coef: 0.0821
Epoch 91/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2825 - mae_euclidean: 0.9451 - jacard_coef: 0.5594 - val_loss: 0.8854 - val_mae_euclidean: 11.5678 - val_jacard_coef: 0.0608
Epoch 92/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2819 - mae_euclidean: 0.9867 - jacard_coef: 0.5603 - val_loss: 0.8284 - val_mae_euclidean: 9.2632 - val_jacard_coef: 0.0939
Epoch 93/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2922 - mae_euclidean: 0.9411 - jacard_coef: 0.5480 - val_loss: 0.8664 - val_mae_euclidean: 9.8835 - val_jacard_coef: 0.0716
Epoch 94/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2870 - mae_euclidean: 1.0717 - jacard_coef: 0.5543 - val_loss: 0.8010 - val_mae_euclidean: 8.5617 - val_jacard_coef: 0.1105
Epoch 95/200
4/4 [==============================] - 2s 398ms/step - loss: 0.3042 - mae_euclidean: 1.2186 - jacard_coef: 0.5358 - val_loss: 0.8011 - val_mae_euclidean: 8.2166 - val_jacard_coef: 0.1104
Epoch 96/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2840 - mae_euclidean: 1.0378 - jacard_coef: 0.5576 - val_loss: 0.7772 - val_mae_euclidean: 8.0867 - val_jacard_coef: 0.1254
Epoch 97/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2853 - mae_euclidean: 1.0830 - jacard_coef: 0.5562 - val_loss: 0.7742 - val_mae_euclidean: 7.1573 - val_jacard_coef: 0.1273
Epoch 98/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2797 - mae_euclidean: 1.0316 - jacard_coef: 0.5630 - val_loss: 0.7634 - val_mae_euclidean: 6.5987 - val_jacard_coef: 0.1342
Epoch 99/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2819 - mae_euclidean: 1.1087 - jacard_coef: 0.5602 - val_loss: 0.7361 - val_mae_euclidean: 6.7281 - val_jacard_coef: 0.1520
Epoch 100/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2736 - mae_euclidean: 0.9992 - jacard_coef: 0.5704 - val_loss: 0.6259 - val_mae_euclidean: 5.0929 - val_jacard_coef: 0.2301
Epoch 101/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2805 - mae_euclidean: 1.1450 - jacard_coef: 0.5621 - val_loss: 0.6335 - val_mae_euclidean: 5.2166 - val_jacard_coef: 0.2244
Epoch 102/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2810 - mae_euclidean: 1.0416 - jacard_coef: 0.5616 - val_loss: 0.6599 - val_mae_euclidean: 5.5884 - val_jacard_coef: 0.2049
Epoch 103/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2738 - mae_euclidean: 0.9386 - jacard_coef: 0.5702 - val_loss: 0.6655 - val_mae_euclidean: 5.1590 - val_jacard_coef: 0.2008
Epoch 104/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2786 - mae_euclidean: 0.9562 - jacard_coef: 0.5644 - val_loss: 0.7641 - val_mae_euclidean: 6.8939 - val_jacard_coef: 0.1337
Epoch 105/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2759 - mae_euclidean: 0.9926 - jacard_coef: 0.5676 - val_loss: 0.5627 - val_mae_euclidean: 3.7714 - val_jacard_coef: 0.2799
Epoch 106/200
4/4 [==============================] - 2s 394ms/step - loss: 0.2771 - mae_euclidean: 1.0808 - jacard_coef: 0.5661 - val_loss: 0.6174 - val_mae_euclidean: 4.0117 - val_jacard_coef: 0.2366
Epoch 107/200
4/4 [==============================] - 2s 395ms/step - loss: 0.2789 - mae_euclidean: 1.0528 - jacard_coef: 0.5639 - val_loss: 0.5794 - val_mae_euclidean: 3.9134 - val_jacard_coef: 0.2663
Epoch 108/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2783 - mae_euclidean: 0.9322 - jacard_coef: 0.5646 - val_loss: 0.4791 - val_mae_euclidean: 3.0845 - val_jacard_coef: 0.3522
Epoch 109/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2837 - mae_euclidean: 1.0240 - jacard_coef: 0.5582 - val_loss: 0.4696 - val_mae_euclidean: 3.0874 - val_jacard_coef: 0.3609
Epoch 110/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2745 - mae_euclidean: 0.8689 - jacard_coef: 0.5694 - val_loss: 0.4929 - val_mae_euclidean: 3.6633 - val_jacard_coef: 0.3397
Epoch 111/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2686 - mae_euclidean: 0.8693 - jacard_coef: 0.5766 - val_loss: 0.4760 - val_mae_euclidean: 3.2610 - val_jacard_coef: 0.3550
Epoch 112/200
4/4 [==============================] - 2s 394ms/step - loss: 0.2817 - mae_euclidean: 1.0345 - jacard_coef: 0.5614 - val_loss: 0.5073 - val_mae_euclidean: 3.4129 - val_jacard_coef: 0.3269
Epoch 113/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2681 - mae_euclidean: 0.9786 - jacard_coef: 0.5773 - val_loss: 0.5021 - val_mae_euclidean: 3.8260 - val_jacard_coef: 0.3315
Epoch 114/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2725 - mae_euclidean: 1.0968 - jacard_coef: 0.5721 - val_loss: 0.5649 - val_mae_euclidean: 4.1720 - val_jacard_coef: 0.2780
Epoch 115/200
4/4 [==============================] - 2s 400ms/step - loss: 0.2692 - mae_euclidean: 0.9822 - jacard_coef: 0.5760 - val_loss: 0.5403 - val_mae_euclidean: 3.9245 - val_jacard_coef: 0.2984
Epoch 116/200
4/4 [==============================] - 2s 400ms/step - loss: 0.2640 - mae_euclidean: 0.8208 - jacard_coef: 0.5826 - val_loss: 0.4533 - val_mae_euclidean: 2.8338 - val_jacard_coef: 0.3762
Epoch 117/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2660 - mae_euclidean: 1.0085 - jacard_coef: 0.5799 - val_loss: 0.4683 - val_mae_euclidean: 3.1867 - val_jacard_coef: 0.3621
Epoch 118/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2584 - mae_euclidean: 0.8705 - jacard_coef: 0.5893 - val_loss: 0.4873 - val_mae_euclidean: 3.7486 - val_jacard_coef: 0.3448
Epoch 119/200
4/4 [==============================] - 2s 394ms/step - loss: 0.2634 - mae_euclidean: 0.8635 - jacard_coef: 0.5831 - val_loss: 0.4614 - val_mae_euclidean: 2.6766 - val_jacard_coef: 0.3686
Epoch 120/200
4/4 [==============================] - 2s 395ms/step - loss: 0.2575 - mae_euclidean: 0.9955 - jacard_coef: 0.5905 - val_loss: 0.4243 - val_mae_euclidean: 2.2947 - val_jacard_coef: 0.4042
Epoch 121/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2593 - mae_euclidean: 0.9961 - jacard_coef: 0.5883 - val_loss: 0.4943 - val_mae_euclidean: 3.3778 - val_jacard_coef: 0.3384
Epoch 122/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2610 - mae_euclidean: 1.0140 - jacard_coef: 0.5863 - val_loss: 0.4653 - val_mae_euclidean: 2.9439 - val_jacard_coef: 0.3649
Epoch 123/200
4/4 [==============================] - 2s 394ms/step - loss: 0.2762 - mae_euclidean: 1.0773 - jacard_coef: 0.5684 - val_loss: 0.4390 - val_mae_euclidean: 2.8095 - val_jacard_coef: 0.3899
Epoch 124/200
4/4 [==============================] - 2s 395ms/step - loss: 0.2615 - mae_euclidean: 0.9738 - jacard_coef: 0.5855 - val_loss: 0.4885 - val_mae_euclidean: 3.3330 - val_jacard_coef: 0.3436
Epoch 125/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2627 - mae_euclidean: 0.8668 - jacard_coef: 0.5842 - val_loss: 0.3926 - val_mae_euclidean: 2.1221 - val_jacard_coef: 0.4362
Epoch 126/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2779 - mae_euclidean: 1.1360 - jacard_coef: 0.5654 - val_loss: 0.4214 - val_mae_euclidean: 2.0887 - val_jacard_coef: 0.4070
Epoch 127/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2687 - mae_euclidean: 0.9210 - jacard_coef: 0.5764 - val_loss: 0.4068 - val_mae_euclidean: 2.3998 - val_jacard_coef: 0.4217
Epoch 128/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2717 - mae_euclidean: 1.2421 - jacard_coef: 0.5729 - val_loss: 0.3762 - val_mae_euclidean: 1.9979 - val_jacard_coef: 0.4533
Epoch 129/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2652 - mae_euclidean: 0.9622 - jacard_coef: 0.5815 - val_loss: 0.3909 - val_mae_euclidean: 2.2906 - val_jacard_coef: 0.4380
Epoch 130/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2653 - mae_euclidean: 0.8393 - jacard_coef: 0.5808 - val_loss: 0.4518 - val_mae_euclidean: 3.2618 - val_jacard_coef: 0.3776
Epoch 131/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2652 - mae_euclidean: 0.8746 - jacard_coef: 0.5809 - val_loss: 0.3734 - val_mae_euclidean: 2.7005 - val_jacard_coef: 0.4563
Epoch 132/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2713 - mae_euclidean: 0.9143 - jacard_coef: 0.5740 - val_loss: 0.3701 - val_mae_euclidean: 2.3283 - val_jacard_coef: 0.4597
Epoch 133/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2565 - mae_euclidean: 0.9569 - jacard_coef: 0.5918 - val_loss: 0.3286 - val_mae_euclidean: 1.2917 - val_jacard_coef: 0.5054
Epoch 134/200
4/4 [==============================] - 2s 394ms/step - loss: 0.2540 - mae_euclidean: 0.8115 - jacard_coef: 0.5951 - val_loss: 0.3902 - val_mae_euclidean: 1.8641 - val_jacard_coef: 0.4387
Epoch 135/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2579 - mae_euclidean: 0.9089 - jacard_coef: 0.5902 - val_loss: 0.3728 - val_mae_euclidean: 1.9669 - val_jacard_coef: 0.4569
Epoch 136/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2557 - mae_euclidean: 0.9829 - jacard_coef: 0.5929 - val_loss: 0.3660 - val_mae_euclidean: 2.2921 - val_jacard_coef: 0.4641
Epoch 137/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2568 - mae_euclidean: 0.9091 - jacard_coef: 0.5916 - val_loss: 0.3671 - val_mae_euclidean: 2.2014 - val_jacard_coef: 0.4630
Epoch 138/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2628 - mae_euclidean: 1.2207 - jacard_coef: 0.5845 - val_loss: 0.4005 - val_mae_euclidean: 2.4907 - val_jacard_coef: 0.4281
Epoch 139/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2525 - mae_euclidean: 0.7571 - jacard_coef: 0.5968 - val_loss: 0.4099 - val_mae_euclidean: 2.9735 - val_jacard_coef: 0.4186
Epoch 140/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2487 - mae_euclidean: 0.7763 - jacard_coef: 0.6018 - val_loss: 0.3909 - val_mae_euclidean: 2.8168 - val_jacard_coef: 0.4380
Epoch 141/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2593 - mae_euclidean: 1.1573 - jacard_coef: 0.5894 - val_loss: 0.3732 - val_mae_euclidean: 2.2009 - val_jacard_coef: 0.4564
Epoch 142/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2610 - mae_euclidean: 0.9138 - jacard_coef: 0.5861 - val_loss: 0.3541 - val_mae_euclidean: 2.0354 - val_jacard_coef: 0.4770
Epoch 143/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2505 - mae_euclidean: 0.8358 - jacard_coef: 0.5994 - val_loss: 0.3197 - val_mae_euclidean: 1.3492 - val_jacard_coef: 0.5156
Epoch 144/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2562 - mae_euclidean: 0.9712 - jacard_coef: 0.5923 - val_loss: 0.3257 - val_mae_euclidean: 1.3358 - val_jacard_coef: 0.5087
Epoch 145/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2513 - mae_euclidean: 0.9045 - jacard_coef: 0.5984 - val_loss: 0.3124 - val_mae_euclidean: 1.2727 - val_jacard_coef: 0.5240
Epoch 146/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2513 - mae_euclidean: 0.8889 - jacard_coef: 0.5985 - val_loss: 0.3379 - val_mae_euclidean: 1.5403 - val_jacard_coef: 0.4949
Epoch 147/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2499 - mae_euclidean: 0.8484 - jacard_coef: 0.6002 - val_loss: 0.3633 - val_mae_euclidean: 1.8823 - val_jacard_coef: 0.4671
Epoch 148/200
4/4 [==============================] - 2s 400ms/step - loss: 0.2511 - mae_euclidean: 0.9882 - jacard_coef: 0.5987 - val_loss: 0.3710 - val_mae_euclidean: 1.9478 - val_jacard_coef: 0.4588
Epoch 149/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2593 - mae_euclidean: 0.8844 - jacard_coef: 0.5884 - val_loss: 0.3731 - val_mae_euclidean: 1.8557 - val_jacard_coef: 0.4566
Epoch 150/200
4/4 [==============================] - 2s 395ms/step - loss: 0.2544 - mae_euclidean: 1.0956 - jacard_coef: 0.5946 - val_loss: 0.3206 - val_mae_euclidean: 1.3182 - val_jacard_coef: 0.5145
Epoch 151/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2506 - mae_euclidean: 0.9059 - jacard_coef: 0.5993 - val_loss: 0.3923 - val_mae_euclidean: 2.1990 - val_jacard_coef: 0.4364
Epoch 152/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2526 - mae_euclidean: 0.8689 - jacard_coef: 0.5970 - val_loss: 0.3544 - val_mae_euclidean: 1.7805 - val_jacard_coef: 0.4767
Epoch 153/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2570 - mae_euclidean: 0.9347 - jacard_coef: 0.5911 - val_loss: 0.3233 - val_mae_euclidean: 1.7045 - val_jacard_coef: 0.5114
Epoch 154/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2610 - mae_euclidean: 0.8848 - jacard_coef: 0.5863 - val_loss: 0.3034 - val_mae_euclidean: 1.1984 - val_jacard_coef: 0.5345
Epoch 155/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2517 - mae_euclidean: 0.8189 - jacard_coef: 0.5978 - val_loss: 0.3454 - val_mae_euclidean: 1.6555 - val_jacard_coef: 0.4866
Epoch 156/200
4/4 [==============================] - 2s 395ms/step - loss: 0.2484 - mae_euclidean: 0.8350 - jacard_coef: 0.6022 - val_loss: 0.3306 - val_mae_euclidean: 1.6669 - val_jacard_coef: 0.5031
Epoch 157/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2470 - mae_euclidean: 0.8744 - jacard_coef: 0.6040 - val_loss: 0.3205 - val_mae_euclidean: 1.2876 - val_jacard_coef: 0.5146
Epoch 158/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2536 - mae_euclidean: 0.9615 - jacard_coef: 0.5955 - val_loss: 0.3886 - val_mae_euclidean: 2.3787 - val_jacard_coef: 0.4403
Epoch 159/200
4/4 [==============================] - 2s 396ms/step - loss: 0.2428 - mae_euclidean: 0.8251 - jacard_coef: 0.6094 - val_loss: 0.3204 - val_mae_euclidean: 1.3421 - val_jacard_coef: 0.5147
Epoch 160/200
4/4 [==============================] - 2s 400ms/step - loss: 0.2420 - mae_euclidean: 0.9169 - jacard_coef: 0.6103 - val_loss: 0.3461 - val_mae_euclidean: 1.4151 - val_jacard_coef: 0.4858
Epoch 161/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2460 - mae_euclidean: 0.8506 - jacard_coef: 0.6051 - val_loss: 0.3244 - val_mae_euclidean: 1.0882 - val_jacard_coef: 0.5101
Epoch 162/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2416 - mae_euclidean: 0.7970 - jacard_coef: 0.6110 - val_loss: 0.3080 - val_mae_euclidean: 1.1912 - val_jacard_coef: 0.5291
Epoch 163/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2391 - mae_euclidean: 0.7884 - jacard_coef: 0.6142 - val_loss: 0.3610 - val_mae_euclidean: 1.6570 - val_jacard_coef: 0.4696
Epoch 164/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2388 - mae_euclidean: 0.8668 - jacard_coef: 0.6144 - val_loss: 0.3269 - val_mae_euclidean: 1.3472 - val_jacard_coef: 0.5073
Epoch 165/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2391 - mae_euclidean: 0.9445 - jacard_coef: 0.6142 - val_loss: 0.3367 - val_mae_euclidean: 1.7319 - val_jacard_coef: 0.4962
Epoch 166/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2373 - mae_euclidean: 0.7719 - jacard_coef: 0.6166 - val_loss: 0.3293 - val_mae_euclidean: 1.5927 - val_jacard_coef: 0.5046
Epoch 167/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2381 - mae_euclidean: 0.8272 - jacard_coef: 0.6156 - val_loss: 0.3265 - val_mae_euclidean: 1.3512 - val_jacard_coef: 0.5078
Epoch 168/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2402 - mae_euclidean: 0.9558 - jacard_coef: 0.6128 - val_loss: 0.3264 - val_mae_euclidean: 1.6288 - val_jacard_coef: 0.5078
Epoch 169/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2374 - mae_euclidean: 0.8358 - jacard_coef: 0.6163 - val_loss: 0.3783 - val_mae_euclidean: 2.0314 - val_jacard_coef: 0.4511
Epoch 170/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2377 - mae_euclidean: 0.8472 - jacard_coef: 0.6159 - val_loss: 0.3244 - val_mae_euclidean: 1.6126 - val_jacard_coef: 0.5101
Epoch 171/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2385 - mae_euclidean: 0.9862 - jacard_coef: 0.6152 - val_loss: 0.3521 - val_mae_euclidean: 1.8797 - val_jacard_coef: 0.4792
Epoch 172/200
4/4 [==============================] - 2s 400ms/step - loss: 0.2541 - mae_euclidean: 1.0677 - jacard_coef: 0.5948 - val_loss: 0.3288 - val_mae_euclidean: 1.2755 - val_jacard_coef: 0.5052
Epoch 173/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2358 - mae_euclidean: 0.8030 - jacard_coef: 0.6185 - val_loss: 0.3513 - val_mae_euclidean: 1.1357 - val_jacard_coef: 0.4801
Epoch 174/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2349 - mae_euclidean: 0.7714 - jacard_coef: 0.6198 - val_loss: 0.3014 - val_mae_euclidean: 1.1567 - val_jacard_coef: 0.5369
Epoch 175/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2342 - mae_euclidean: 0.8019 - jacard_coef: 0.6206 - val_loss: 0.3204 - val_mae_euclidean: 1.2285 - val_jacard_coef: 0.5147
Epoch 176/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2410 - mae_euclidean: 1.0110 - jacard_coef: 0.6120 - val_loss: 0.3057 - val_mae_euclidean: 1.1025 - val_jacard_coef: 0.5317
Epoch 177/200
4/4 [==============================] - 2s 402ms/step - loss: 0.2415 - mae_euclidean: 0.7881 - jacard_coef: 0.6110 - val_loss: 0.3246 - val_mae_euclidean: 1.2411 - val_jacard_coef: 0.5099
Epoch 178/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2348 - mae_euclidean: 1.0649 - jacard_coef: 0.6198 - val_loss: 0.3039 - val_mae_euclidean: 1.0200 - val_jacard_coef: 0.5338
Epoch 179/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2350 - mae_euclidean: 0.8386 - jacard_coef: 0.6197 - val_loss: 0.2999 - val_mae_euclidean: 1.0664 - val_jacard_coef: 0.5385
Epoch 180/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2362 - mae_euclidean: 0.8652 - jacard_coef: 0.6182 - val_loss: 0.3076 - val_mae_euclidean: 1.2001 - val_jacard_coef: 0.5295
Epoch 181/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2392 - mae_euclidean: 0.8280 - jacard_coef: 0.6141 - val_loss: 0.3293 - val_mae_euclidean: 1.7501 - val_jacard_coef: 0.5045
Epoch 182/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2361 - mae_euclidean: 0.8030 - jacard_coef: 0.6183 - val_loss: 0.3101 - val_mae_euclidean: 1.3649 - val_jacard_coef: 0.5266
Epoch 183/200
4/4 [==============================] - 2s 400ms/step - loss: 0.2474 - mae_euclidean: 0.8315 - jacard_coef: 0.6036 - val_loss: 0.3355 - val_mae_euclidean: 1.3453 - val_jacard_coef: 0.4976
Epoch 184/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2388 - mae_euclidean: 0.7858 - jacard_coef: 0.6147 - val_loss: 0.3346 - val_mae_euclidean: 1.8751 - val_jacard_coef: 0.4986
Epoch 185/200
4/4 [==============================] - 2s 398ms/step - loss: 0.2331 - mae_euclidean: 0.7514 - jacard_coef: 0.6220 - val_loss: 0.2983 - val_mae_euclidean: 1.2966 - val_jacard_coef: 0.5405
Epoch 186/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2339 - mae_euclidean: 0.8246 - jacard_coef: 0.6210 - val_loss: 0.3048 - val_mae_euclidean: 1.2117 - val_jacard_coef: 0.5328
Epoch 187/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2347 - mae_euclidean: 0.9667 - jacard_coef: 0.6199 - val_loss: 0.3003 - val_mae_euclidean: 1.0822 - val_jacard_coef: 0.5381
Epoch 188/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2362 - mae_euclidean: 0.8339 - jacard_coef: 0.6180 - val_loss: 0.3050 - val_mae_euclidean: 1.3327 - val_jacard_coef: 0.5325
Epoch 189/200
4/4 [==============================] - 2s 397ms/step - loss: 0.2321 - mae_euclidean: 0.8541 - jacard_coef: 0.6234 - val_loss: 0.3115 - val_mae_euclidean: 1.2849 - val_jacard_coef: 0.5250
Epoch 190/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2238 - mae_euclidean: 0.7100 - jacard_coef: 0.6344 - val_loss: 0.3100 - val_mae_euclidean: 1.0555 - val_jacard_coef: 0.5267
Epoch 191/200
4/4 [==============================] - 2s 403ms/step - loss: 0.2239 - mae_euclidean: 0.7549 - jacard_coef: 0.6341 - val_loss: 0.3254 - val_mae_euclidean: 1.2957 - val_jacard_coef: 0.5090
Epoch 192/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2274 - mae_euclidean: 0.7543 - jacard_coef: 0.6296 - val_loss: 0.2995 - val_mae_euclidean: 0.9952 - val_jacard_coef: 0.5391
Epoch 193/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2264 - mae_euclidean: 0.7329 - jacard_coef: 0.6308 - val_loss: 0.3195 - val_mae_euclidean: 1.1394 - val_jacard_coef: 0.5158
Epoch 194/200
4/4 [==============================] - 2s 400ms/step - loss: 0.2304 - mae_euclidean: 0.9172 - jacard_coef: 0.6257 - val_loss: 0.3028 - val_mae_euclidean: 0.8840 - val_jacard_coef: 0.5351
Epoch 195/200
4/4 [==============================] - 2s 403ms/step - loss: 0.2469 - mae_euclidean: 1.2037 - jacard_coef: 0.6045 - val_loss: 0.3098 - val_mae_euclidean: 1.1743 - val_jacard_coef: 0.5269
Epoch 196/200
4/4 [==============================] - 2s 401ms/step - loss: 0.2445 - mae_euclidean: 0.9682 - jacard_coef: 0.6074 - val_loss: 0.3127 - val_mae_euclidean: 1.3041 - val_jacard_coef: 0.5236
Epoch 197/200
4/4 [==============================] - 2s 402ms/step - loss: 0.2391 - mae_euclidean: 1.0908 - jacard_coef: 0.6144 - val_loss: 0.3300 - val_mae_euclidean: 1.4195 - val_jacard_coef: 0.5037
Epoch 198/200
4/4 [==============================] - 2s 402ms/step - loss: 0.2315 - mae_euclidean: 1.0427 - jacard_coef: 0.6242 - val_loss: 0.3029 - val_mae_euclidean: 0.9921 - val_jacard_coef: 0.5351
Epoch 199/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2250 - mae_euclidean: 0.8276 - jacard_coef: 0.6328 - val_loss: 0.2970 - val_mae_euclidean: 1.2448 - val_jacard_coef: 0.5420
Epoch 200/200
4/4 [==============================] - 2s 399ms/step - loss: 0.2276 - mae_euclidean: 0.8272 - jacard_coef: 0.6293 - val_loss: 0.2979 - val_mae_euclidean: 1.2925 - val_jacard_coef: 0.5409
Model trained for 316.5363738536835s

C:\Users\a1048794\Anaconda3\lib\site-packages\keras\engine\functional.py:1410: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.
  layer_config = serialize_layer_fn(layer)

plot_history(history_sa_unet1, "SA Unet")

model_sa_unet1.evaluate(X_val, y_val)

1/1 [==============================] - 0s 353ms/step - loss: 0.2979 - mae_euclidean: 1.2925 - jacard_coef: 0.5409

[0.2979312539100647, 1.292516827583313, 0.540920078754425]

Attention UNet model trials

# tf.keras.backend.clear_session()

# Attention Unet1
tf.random.set_seed(14)
model_att_unet1 = Attention_UNet(input_shape)
model_att_unet1.compile(optimizer=Adam(learning_rate=1e-2),
                         loss = dice_coef_loss,
                         metrics=[mae_euclidean, jacard_coef]
                         )
start = time.time()
history_att_unet1 = model_att_unet1.fit(X_train, y_train,
                                    validation_data = (X_val, y_val),
                                    batch_size = 2,
                                    epochs=50
                                   )
print(f"Model trained for {time.time() - start}s")
model_att_unet1.save("2022-02-17 Att-UNet1 50epochs.hdf5")

Epoch 1/50
16/16 [==============================] - 19s 455ms/step - loss: 0.8100 - mae_euclidean: 4.1651 - jacard_coef: 0.1062 - val_loss: 0.9816 - val_mae_euclidean: 63.0797 - val_jacard_coef: 0.0093
Epoch 2/50
16/16 [==============================] - 7s 418ms/step - loss: 0.7642 - mae_euclidean: 3.8893 - jacard_coef: 0.1349 - val_loss: 0.8609 - val_mae_euclidean: 6.2517 - val_jacard_coef: 0.0759
Epoch 3/50
16/16 [==============================] - 7s 419ms/step - loss: 0.7339 - mae_euclidean: 3.7794 - jacard_coef: 0.1547 - val_loss: 0.9324 - val_mae_euclidean: 57.5636 - val_jacard_coef: 0.0351
Epoch 4/50
16/16 [==============================] - 7s 419ms/step - loss: 0.7081 - mae_euclidean: 3.0676 - jacard_coef: 0.1723 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.8732e-04
Epoch 5/50
16/16 [==============================] - 7s 418ms/step - loss: 0.6804 - mae_euclidean: 2.9842 - jacard_coef: 0.1919 - val_loss: 0.8808 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 0.0638
Epoch 6/50
16/16 [==============================] - 7s 419ms/step - loss: 0.6518 - mae_euclidean: 2.7565 - jacard_coef: 0.2140 - val_loss: 0.9727 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 0.0139
Epoch 7/50
16/16 [==============================] - 7s 420ms/step - loss: 0.6214 - mae_euclidean: 2.3451 - jacard_coef: 0.2370 - val_loss: 0.8643 - val_mae_euclidean: 46.9861 - val_jacard_coef: 0.0736
Epoch 8/50
16/16 [==============================] - 7s 419ms/step - loss: 0.5887 - mae_euclidean: 2.1963 - jacard_coef: 0.2613 - val_loss: 0.8825 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0628
Epoch 9/50
16/16 [==============================] - 7s 422ms/step - loss: 0.5743 - mae_euclidean: 2.3518 - jacard_coef: 0.2745 - val_loss: 0.9922 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0040
Epoch 10/50
16/16 [==============================] - 7s 421ms/step - loss: 0.5328 - mae_euclidean: 1.8690 - jacard_coef: 0.3073 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.1589e-04
Epoch 11/50
16/16 [==============================] - 7s 421ms/step - loss: 0.5231 - mae_euclidean: 1.9893 - jacard_coef: 0.3175 - val_loss: 0.8616 - val_mae_euclidean: 6.3162 - val_jacard_coef: 0.0755
Epoch 12/50
16/16 [==============================] - 7s 422ms/step - loss: 0.5009 - mae_euclidean: 1.9681 - jacard_coef: 0.3369 - val_loss: 0.7520 - val_mae_euclidean: 5.7388 - val_jacard_coef: 0.1441
Epoch 13/50
16/16 [==============================] - 7s 421ms/step - loss: 0.4705 - mae_euclidean: 1.4292 - jacard_coef: 0.3637 - val_loss: 0.8644 - val_mae_euclidean: 11.1895 - val_jacard_coef: 0.0736
Epoch 14/50
16/16 [==============================] - 7s 420ms/step - loss: 0.4530 - mae_euclidean: 1.5741 - jacard_coef: 0.3800 - val_loss: 0.8470 - val_mae_euclidean: 10.3622 - val_jacard_coef: 0.0840
Epoch 15/50
16/16 [==============================] - 7s 418ms/step - loss: 0.4277 - mae_euclidean: 1.5537 - jacard_coef: 0.4033 - val_loss: 0.8496 - val_mae_euclidean: 11.0072 - val_jacard_coef: 0.0822
Epoch 16/50
16/16 [==============================] - 7s 418ms/step - loss: 0.4163 - mae_euclidean: 1.2416 - jacard_coef: 0.4148 - val_loss: 0.8369 - val_mae_euclidean: 26.9060 - val_jacard_coef: 0.0926
Epoch 17/50
16/16 [==============================] - 7s 415ms/step - loss: 0.4033 - mae_euclidean: 1.4298 - jacard_coef: 0.4280 - val_loss: 0.9986 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.0801e-04
Epoch 18/50
16/16 [==============================] - 7s 416ms/step - loss: 0.4008 - mae_euclidean: 1.2758 - jacard_coef: 0.4300 - val_loss: 0.7208 - val_mae_euclidean: 7.6967 - val_jacard_coef: 0.1660
Epoch 19/50
16/16 [==============================] - 7s 416ms/step - loss: 0.3982 - mae_euclidean: 1.6878 - jacard_coef: 0.4341 - val_loss: 0.6972 - val_mae_euclidean: 5.2524 - val_jacard_coef: 0.1828
Epoch 20/50
16/16 [==============================] - 7s 415ms/step - loss: 0.3826 - mae_euclidean: 1.2828 - jacard_coef: 0.4486 - val_loss: 0.7442 - val_mae_euclidean: 6.2267 - val_jacard_coef: 0.1487
Epoch 21/50
16/16 [==============================] - 7s 415ms/step - loss: 0.3743 - mae_euclidean: 1.1040 - jacard_coef: 0.4577 - val_loss: 0.6309 - val_mae_euclidean: 4.6526 - val_jacard_coef: 0.2338
Epoch 22/50
16/16 [==============================] - 7s 416ms/step - loss: 0.3643 - mae_euclidean: 1.3188 - jacard_coef: 0.4689 - val_loss: 0.6393 - val_mae_euclidean: 3.5176 - val_jacard_coef: 0.2248
Epoch 23/50
16/16 [==============================] - 7s 418ms/step - loss: 0.3664 - mae_euclidean: 1.0332 - jacard_coef: 0.4659 - val_loss: 0.6965 - val_mae_euclidean: 8.9686 - val_jacard_coef: 0.1855
Epoch 24/50
16/16 [==============================] - 7s 419ms/step - loss: 0.3462 - mae_euclidean: 1.0379 - jacard_coef: 0.4883 - val_loss: 0.8356 - val_mae_euclidean: 14.9888 - val_jacard_coef: 0.0912
Epoch 25/50
16/16 [==============================] - 7s 420ms/step - loss: 0.3313 - mae_euclidean: 1.0257 - jacard_coef: 0.5041 - val_loss: 0.6442 - val_mae_euclidean: 3.5344 - val_jacard_coef: 0.2326
Epoch 26/50
16/16 [==============================] - 7s 422ms/step - loss: 0.3291 - mae_euclidean: 1.0188 - jacard_coef: 0.5064 - val_loss: 0.4983 - val_mae_euclidean: 4.0402 - val_jacard_coef: 0.3401
Epoch 27/50
16/16 [==============================] - 7s 421ms/step - loss: 0.3368 - mae_euclidean: 1.0869 - jacard_coef: 0.4984 - val_loss: 0.4648 - val_mae_euclidean: 2.6472 - val_jacard_coef: 0.3704
Epoch 28/50
16/16 [==============================] - 7s 422ms/step - loss: 0.3384 - mae_euclidean: 1.0450 - jacard_coef: 0.4975 - val_loss: 0.5649 - val_mae_euclidean: 3.9019 - val_jacard_coef: 0.2809
Epoch 29/50
16/16 [==============================] - 7s 420ms/step - loss: 0.3244 - mae_euclidean: 1.0500 - jacard_coef: 0.5123 - val_loss: 0.4317 - val_mae_euclidean: 2.1008 - val_jacard_coef: 0.4006
Epoch 30/50
16/16 [==============================] - 7s 420ms/step - loss: 0.3072 - mae_euclidean: 0.9288 - jacard_coef: 0.5317 - val_loss: 0.4683 - val_mae_euclidean: 2.6413 - val_jacard_coef: 0.3656
Epoch 31/50
16/16 [==============================] - 7s 424ms/step - loss: 0.3118 - mae_euclidean: 0.9120 - jacard_coef: 0.5273 - val_loss: 0.4665 - val_mae_euclidean: 3.2884 - val_jacard_coef: 0.3688
Epoch 32/50
16/16 [==============================] - 7s 421ms/step - loss: 0.3044 - mae_euclidean: 0.8473 - jacard_coef: 0.5352 - val_loss: 0.4244 - val_mae_euclidean: 2.4463 - val_jacard_coef: 0.4088
Epoch 33/50
16/16 [==============================] - 7s 422ms/step - loss: 0.2948 - mae_euclidean: 0.8533 - jacard_coef: 0.5455 - val_loss: 0.4399 - val_mae_euclidean: 2.6703 - val_jacard_coef: 0.3932
Epoch 34/50
16/16 [==============================] - 7s 421ms/step - loss: 0.2881 - mae_euclidean: 0.8028 - jacard_coef: 0.5540 - val_loss: 0.5154 - val_mae_euclidean: 3.5347 - val_jacard_coef: 0.3210
Epoch 35/50
16/16 [==============================] - 7s 422ms/step - loss: 0.3180 - mae_euclidean: 1.0144 - jacard_coef: 0.5196 - val_loss: 0.3878 - val_mae_euclidean: 2.1736 - val_jacard_coef: 0.4428
Epoch 36/50
16/16 [==============================] - 7s 422ms/step - loss: 0.3116 - mae_euclidean: 1.1857 - jacard_coef: 0.5273 - val_loss: 0.3779 - val_mae_euclidean: 1.5760 - val_jacard_coef: 0.4540
Epoch 37/50
16/16 [==============================] - 7s 421ms/step - loss: 0.2887 - mae_euclidean: 0.8191 - jacard_coef: 0.5538 - val_loss: 0.3493 - val_mae_euclidean: 1.3364 - val_jacard_coef: 0.4842
Epoch 38/50
16/16 [==============================] - 7s 422ms/step - loss: 0.2879 - mae_euclidean: 0.8351 - jacard_coef: 0.5550 - val_loss: 0.3724 - val_mae_euclidean: 2.0884 - val_jacard_coef: 0.4595
Epoch 39/50
16/16 [==============================] - 7s 422ms/step - loss: 0.2763 - mae_euclidean: 0.7798 - jacard_coef: 0.5688 - val_loss: 0.3388 - val_mae_euclidean: 1.1190 - val_jacard_coef: 0.4954
Epoch 40/50
16/16 [==============================] - 7s 422ms/step - loss: 0.2610 - mae_euclidean: 0.7614 - jacard_coef: 0.5872 - val_loss: 0.3519 - val_mae_euclidean: 1.6819 - val_jacard_coef: 0.4809
Epoch 41/50
16/16 [==============================] - 7s 421ms/step - loss: 0.2694 - mae_euclidean: 0.7487 - jacard_coef: 0.5773 - val_loss: 0.3585 - val_mae_euclidean: 2.3875 - val_jacard_coef: 0.4733
Epoch 42/50
16/16 [==============================] - 7s 422ms/step - loss: 0.2782 - mae_euclidean: 0.7625 - jacard_coef: 0.5658 - val_loss: 0.4312 - val_mae_euclidean: 3.0479 - val_jacard_coef: 0.3982
Epoch 43/50
16/16 [==============================] - 7s 423ms/step - loss: 0.2548 - mae_euclidean: 0.8217 - jacard_coef: 0.5949 - val_loss: 0.3958 - val_mae_euclidean: 2.0207 - val_jacard_coef: 0.4332
Epoch 44/50
16/16 [==============================] - 7s 422ms/step - loss: 0.2557 - mae_euclidean: 0.6456 - jacard_coef: 0.5945 - val_loss: 0.3771 - val_mae_euclidean: 1.4731 - val_jacard_coef: 0.4542
Epoch 45/50
16/16 [==============================] - 7s 423ms/step - loss: 0.2494 - mae_euclidean: 0.6942 - jacard_coef: 0.6019 - val_loss: 0.3663 - val_mae_euclidean: 1.3279 - val_jacard_coef: 0.4660
Epoch 46/50
16/16 [==============================] - 7s 422ms/step - loss: 0.2599 - mae_euclidean: 0.6731 - jacard_coef: 0.5893 - val_loss: 0.3355 - val_mae_euclidean: 1.2177 - val_jacard_coef: 0.4994
Epoch 47/50
16/16 [==============================] - 7s 420ms/step - loss: 0.2485 - mae_euclidean: 0.6693 - jacard_coef: 0.6029 - val_loss: 0.3397 - val_mae_euclidean: 1.2132 - val_jacard_coef: 0.4945
Epoch 48/50
16/16 [==============================] - 7s 420ms/step - loss: 0.2569 - mae_euclidean: 0.7965 - jacard_coef: 0.5932 - val_loss: 0.4101 - val_mae_euclidean: 1.5252 - val_jacard_coef: 0.4190
Epoch 49/50
16/16 [==============================] - 7s 420ms/step - loss: 0.2267 - mae_euclidean: 0.6230 - jacard_coef: 0.6309 - val_loss: 0.3851 - val_mae_euclidean: 1.3435 - val_jacard_coef: 0.4451
Epoch 50/50
16/16 [==============================] - 7s 420ms/step - loss: 0.2231 - mae_euclidean: 0.6283 - jacard_coef: 0.6359 - val_loss: 0.3376 - val_mae_euclidean: 1.2765 - val_jacard_coef: 0.4964
Model trained for 347.18052101135254s

plot_history(history_att_unet1, "Attention Unet")

# additional trainning with smaller learning rate
model_att_unet1.compile(optimizer=Adam(learning_rate=1e-3),
                         loss = dice_coef_loss,
                         metrics=[mae_euclidean, jacard_coef]
                         )
start = time.time()
history_att_unet1a = model_att_unet1.fit(X_train, y_train,
                                        validation_data = (X_val, y_val),
                                        batch_size = 2,
                                        epochs=25
                                       )
print(f"Model trained for {time.time() - start}s")
model_att_unet1.save("2022-02-17 Att-UNet1a 50+25epochs.hdf5")

Epoch 1/25
16/16 [==============================] - 11s 453ms/step - loss: 0.2119 - mae_euclidean: 0.5784 - jacard_coef: 0.6516 - val_loss: 0.3437 - val_mae_euclidean: 1.2986 - val_jacard_coef: 0.4897
Epoch 2/25
16/16 [==============================] - 7s 415ms/step - loss: 0.1986 - mae_euclidean: 0.5058 - jacard_coef: 0.6698 - val_loss: 0.3321 - val_mae_euclidean: 1.2732 - val_jacard_coef: 0.5026
Epoch 3/25
16/16 [==============================] - 7s 417ms/step - loss: 0.1919 - mae_euclidean: 0.5179 - jacard_coef: 0.6790 - val_loss: 0.3273 - val_mae_euclidean: 1.2761 - val_jacard_coef: 0.5082
Epoch 4/25
16/16 [==============================] - 7s 419ms/step - loss: 0.1882 - mae_euclidean: 0.5066 - jacard_coef: 0.6842 - val_loss: 0.3243 - val_mae_euclidean: 1.1886 - val_jacard_coef: 0.5121
Epoch 5/25
16/16 [==============================] - 7s 419ms/step - loss: 0.1818 - mae_euclidean: 0.4620 - jacard_coef: 0.6929 - val_loss: 0.3299 - val_mae_euclidean: 1.2154 - val_jacard_coef: 0.5055
Epoch 6/25
16/16 [==============================] - 7s 419ms/step - loss: 0.1839 - mae_euclidean: 0.4857 - jacard_coef: 0.6902 - val_loss: 0.3281 - val_mae_euclidean: 1.1718 - val_jacard_coef: 0.5078
Epoch 7/25
16/16 [==============================] - 7s 418ms/step - loss: 0.1838 - mae_euclidean: 0.4634 - jacard_coef: 0.6907 - val_loss: 0.3285 - val_mae_euclidean: 1.1783 - val_jacard_coef: 0.5073
Epoch 8/25
16/16 [==============================] - 7s 418ms/step - loss: 0.1746 - mae_euclidean: 0.4565 - jacard_coef: 0.7035 - val_loss: 0.3233 - val_mae_euclidean: 1.1359 - val_jacard_coef: 0.5133
Epoch 9/25
16/16 [==============================] - 7s 420ms/step - loss: 0.1786 - mae_euclidean: 0.4360 - jacard_coef: 0.6979 - val_loss: 0.3256 - val_mae_euclidean: 1.0935 - val_jacard_coef: 0.5110
Epoch 10/25
16/16 [==============================] - 7s 421ms/step - loss: 0.1689 - mae_euclidean: 0.4506 - jacard_coef: 0.7115 - val_loss: 0.3277 - val_mae_euclidean: 1.1710 - val_jacard_coef: 0.5086
Epoch 11/25
16/16 [==============================] - 7s 420ms/step - loss: 0.1716 - mae_euclidean: 0.4184 - jacard_coef: 0.7079 - val_loss: 0.3273 - val_mae_euclidean: 1.1730 - val_jacard_coef: 0.5090
Epoch 12/25
16/16 [==============================] - 7s 424ms/step - loss: 0.1668 - mae_euclidean: 0.4064 - jacard_coef: 0.7148 - val_loss: 0.3276 - val_mae_euclidean: 1.1616 - val_jacard_coef: 0.5091
Epoch 13/25
16/16 [==============================] - 7s 437ms/step - loss: 0.1658 - mae_euclidean: 0.4019 - jacard_coef: 0.7164 - val_loss: 0.3226 - val_mae_euclidean: 1.0921 - val_jacard_coef: 0.5140
Epoch 14/25
16/16 [==============================] - 7s 433ms/step - loss: 0.1636 - mae_euclidean: 0.4279 - jacard_coef: 0.7196 - val_loss: 0.3250 - val_mae_euclidean: 1.1222 - val_jacard_coef: 0.5113
Epoch 15/25
16/16 [==============================] - 7s 428ms/step - loss: 0.1594 - mae_euclidean: 0.3838 - jacard_coef: 0.7256 - val_loss: 0.3259 - val_mae_euclidean: 1.1865 - val_jacard_coef: 0.5108
Epoch 16/25
16/16 [==============================] - 7s 423ms/step - loss: 0.1563 - mae_euclidean: 0.3965 - jacard_coef: 0.7303 - val_loss: 0.3243 - val_mae_euclidean: 1.2000 - val_jacard_coef: 0.5126
Epoch 17/25
16/16 [==============================] - 7s 420ms/step - loss: 0.1562 - mae_euclidean: 0.4186 - jacard_coef: 0.7304 - val_loss: 0.3274 - val_mae_euclidean: 1.2233 - val_jacard_coef: 0.5092
Epoch 18/25
16/16 [==============================] - 7s 419ms/step - loss: 0.1524 - mae_euclidean: 0.4166 - jacard_coef: 0.7362 - val_loss: 0.3206 - val_mae_euclidean: 1.1105 - val_jacard_coef: 0.5166
Epoch 19/25
16/16 [==============================] - 7s 418ms/step - loss: 0.1519 - mae_euclidean: 0.3887 - jacard_coef: 0.7368 - val_loss: 0.3273 - val_mae_euclidean: 1.1641 - val_jacard_coef: 0.5093
Epoch 20/25
16/16 [==============================] - 7s 417ms/step - loss: 0.1462 - mae_euclidean: 0.3764 - jacard_coef: 0.7453 - val_loss: 0.3274 - val_mae_euclidean: 1.1778 - val_jacard_coef: 0.5094
Epoch 21/25
16/16 [==============================] - 7s 411ms/step - loss: 0.1479 - mae_euclidean: 0.3788 - jacard_coef: 0.7429 - val_loss: 0.3239 - val_mae_euclidean: 1.1453 - val_jacard_coef: 0.5132
Epoch 22/25
16/16 [==============================] - 7s 408ms/step - loss: 0.1427 - mae_euclidean: 0.3457 - jacard_coef: 0.7508 - val_loss: 0.3273 - val_mae_euclidean: 1.1438 - val_jacard_coef: 0.5092
Epoch 23/25
16/16 [==============================] - 7s 409ms/step - loss: 0.1431 - mae_euclidean: 0.3631 - jacard_coef: 0.7501 - val_loss: 0.3264 - val_mae_euclidean: 1.1133 - val_jacard_coef: 0.5099
Epoch 24/25
16/16 [==============================] - 7s 409ms/step - loss: 0.1392 - mae_euclidean: 0.3698 - jacard_coef: 0.7561 - val_loss: 0.3241 - val_mae_euclidean: 1.1641 - val_jacard_coef: 0.5126
Epoch 25/25
16/16 [==============================] - 6s 407ms/step - loss: 0.1367 - mae_euclidean: 0.3588 - jacard_coef: 0.7601 - val_loss: 0.3246 - val_mae_euclidean: 1.1399 - val_jacard_coef: 0.5120
Model trained for 171.23888444900513s

plot_history(history_att_unet1a, "Att Unet additional trainning")

model_att_unet1.evaluate(X_val, y_val)

1/1 [==============================] - 7s 7s/step - loss: 0.3036 - mae_euclidean: 1.1399 - jacard_coef: 0.5342

[0.3036247491836548, 1.139896035194397, 0.5341904759407043]

UNet (classic) model trials

# tf.keras.backend.clear_session()

# Classic Unet1
tf.random.set_seed(14)
model_unet1 = UNet(input_shape)
model_unet1.compile(optimizer=Adam(learning_rate=1e-2),
                     loss = dice_coef_loss,
                     metrics=[mae_euclidean, jacard_coef]
                     )
start = time.time()
history_unet1 = model_unet1.fit(X_train, y_train,
                                validation_data = (X_val, y_val),
                                batch_size = 4,
                                epochs=200
                               )
print(f"Model trained for {time.time() - start}s")
model_unet1.save("2022-02-17 UNet1 200epochs.hdf5")

Epoch 1/200
8/8 [==============================] - 13s 689ms/step - loss: 0.8266 - mae_euclidean: 4.5184 - jacard_coef: 0.0958 - val_loss: 0.8604 - val_mae_euclidean: 6.2757 - val_jacard_coef: 0.0761
Epoch 2/200
8/8 [==============================] - 5s 617ms/step - loss: 0.7856 - mae_euclidean: 3.6526 - jacard_coef: 0.1209 - val_loss: 0.8619 - val_mae_euclidean: 6.3189 - val_jacard_coef: 0.0753
Epoch 3/200
8/8 [==============================] - 5s 614ms/step - loss: 0.7602 - mae_euclidean: 3.3592 - jacard_coef: 0.1364 - val_loss: 0.8619 - val_mae_euclidean: 6.3222 - val_jacard_coef: 0.0753
Epoch 4/200
8/8 [==============================] - 5s 617ms/step - loss: 0.7461 - mae_euclidean: 3.4805 - jacard_coef: 0.1458 - val_loss: 0.9373 - val_mae_euclidean: 56.7043 - val_jacard_coef: 0.0326
Epoch 5/200
8/8 [==============================] - 5s 617ms/step - loss: 0.7336 - mae_euclidean: 3.3536 - jacard_coef: 0.1546 - val_loss: 0.9834 - val_mae_euclidean: 62.6911 - val_jacard_coef: 0.0084
Epoch 6/200
8/8 [==============================] - 5s 616ms/step - loss: 0.7218 - mae_euclidean: 3.4944 - jacard_coef: 0.1632 - val_loss: 0.9965 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0018
Epoch 7/200
8/8 [==============================] - 5s 614ms/step - loss: 0.7044 - mae_euclidean: 3.0575 - jacard_coef: 0.1748 - val_loss: 0.9668 - val_mae_euclidean: 43.1846 - val_jacard_coef: 0.0170
Epoch 8/200
8/8 [==============================] - 5s 616ms/step - loss: 0.6867 - mae_euclidean: 2.8729 - jacard_coef: 0.1869 - val_loss: 0.8828 - val_mae_euclidean: 6917528477885267968.0000 - val_jacard_coef: 0.0625
Epoch 9/200
8/8 [==============================] - 5s 616ms/step - loss: 0.6800 - mae_euclidean: 2.7929 - jacard_coef: 0.1924 - val_loss: 0.9488 - val_mae_euclidean: 79.7474 - val_jacard_coef: 0.0263
Epoch 10/200
8/8 [==============================] - 5s 615ms/step - loss: 0.6543 - mae_euclidean: 2.4537 - jacard_coef: 0.2097 - val_loss: 0.8906 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 0.0582
Epoch 11/200
8/8 [==============================] - 5s 616ms/step - loss: 0.6454 - mae_euclidean: 2.5071 - jacard_coef: 0.2179 - val_loss: 0.8957 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 0.0553
Epoch 12/200
8/8 [==============================] - 5s 616ms/step - loss: 0.6190 - mae_euclidean: 2.3996 - jacard_coef: 0.2361 - val_loss: 0.9022 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 0.0517
Epoch 13/200
8/8 [==============================] - 5s 616ms/step - loss: 0.6142 - mae_euclidean: 2.2077 - jacard_coef: 0.2407 - val_loss: 0.8736 - val_mae_euclidean: 28.7642 - val_jacard_coef: 0.0681
Epoch 14/200
8/8 [==============================] - 5s 616ms/step - loss: 0.5975 - mae_euclidean: 2.1766 - jacard_coef: 0.2539 - val_loss: 0.9451 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 0.0283
Epoch 15/200
8/8 [==============================] - 5s 614ms/step - loss: 0.5729 - mae_euclidean: 2.0772 - jacard_coef: 0.2723 - val_loss: 0.9203 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 0.0417
Epoch 16/200
8/8 [==============================] - 5s 613ms/step - loss: 0.5632 - mae_euclidean: 2.0378 - jacard_coef: 0.2808 - val_loss: 0.9180 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0430
Epoch 17/200
8/8 [==============================] - 5s 613ms/step - loss: 0.5457 - mae_euclidean: 1.8294 - jacard_coef: 0.2945 - val_loss: 0.9230 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 0.0401
Epoch 18/200
8/8 [==============================] - 5s 611ms/step - loss: 0.5322 - mae_euclidean: 1.8024 - jacard_coef: 0.3072 - val_loss: 0.9277 - val_mae_euclidean: 13835056955770535936.0000 - val_jacard_coef: 0.0376
Epoch 19/200
8/8 [==============================] - 5s 612ms/step - loss: 0.5234 - mae_euclidean: 2.0947 - jacard_coef: 0.3155 - val_loss: 0.9293 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 0.0368
Epoch 20/200
8/8 [==============================] - 5s 613ms/step - loss: 0.5140 - mae_euclidean: 1.6603 - jacard_coef: 0.3229 - val_loss: 0.9304 - val_mae_euclidean: 57.2193 - val_jacard_coef: 0.0362
Epoch 21/200
8/8 [==============================] - 5s 612ms/step - loss: 0.5026 - mae_euclidean: 1.6695 - jacard_coef: 0.3331 - val_loss: 0.9323 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0351
Epoch 22/200
8/8 [==============================] - 5s 609ms/step - loss: 0.4852 - mae_euclidean: 1.6356 - jacard_coef: 0.3495 - val_loss: 0.9322 - val_mae_euclidean: 58.2939 - val_jacard_coef: 0.0352
Epoch 23/200
8/8 [==============================] - 5s 609ms/step - loss: 0.4768 - mae_euclidean: 1.6417 - jacard_coef: 0.3572 - val_loss: 0.9396 - val_mae_euclidean: 6917528477885267968.0000 - val_jacard_coef: 0.0312
Epoch 24/200
8/8 [==============================] - 5s 610ms/step - loss: 0.4501 - mae_euclidean: 1.3466 - jacard_coef: 0.3798 - val_loss: 0.9438 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0290
Epoch 25/200
8/8 [==============================] - 5s 612ms/step - loss: 0.4425 - mae_euclidean: 1.4457 - jacard_coef: 0.3875 - val_loss: 0.9435 - val_mae_euclidean: 57.6788 - val_jacard_coef: 0.0291
Epoch 26/200
8/8 [==============================] - 5s 612ms/step - loss: 0.4302 - mae_euclidean: 1.2779 - jacard_coef: 0.3994 - val_loss: 0.9467 - val_mae_euclidean: 52.1623 - val_jacard_coef: 0.0274
Epoch 27/200
8/8 [==============================] - 5s 611ms/step - loss: 0.4294 - mae_euclidean: 1.5023 - jacard_coef: 0.4031 - val_loss: 0.9487 - val_mae_euclidean: 52.1767 - val_jacard_coef: 0.0263
Epoch 28/200
8/8 [==============================] - 5s 611ms/step - loss: 0.4493 - mae_euclidean: 1.8342 - jacard_coef: 0.3835 - val_loss: 0.9129 - val_mae_euclidean: 21.1828 - val_jacard_coef: 0.0455
Epoch 29/200
8/8 [==============================] - 5s 610ms/step - loss: 0.4194 - mae_euclidean: 1.1538 - jacard_coef: 0.4095 - val_loss: 0.9531 - val_mae_euclidean: 48.6527 - val_jacard_coef: 0.0240
Epoch 30/200
8/8 [==============================] - 5s 611ms/step - loss: 0.4155 - mae_euclidean: 1.2992 - jacard_coef: 0.4144 - val_loss: 0.9425 - val_mae_euclidean: 39.7856 - val_jacard_coef: 0.0296
Epoch 31/200
8/8 [==============================] - 5s 607ms/step - loss: 0.4030 - mae_euclidean: 1.1558 - jacard_coef: 0.4263 - val_loss: 0.9350 - val_mae_euclidean: 18.5468 - val_jacard_coef: 0.0336
Epoch 32/200
8/8 [==============================] - 5s 607ms/step - loss: 0.4041 - mae_euclidean: 1.3065 - jacard_coef: 0.4259 - val_loss: 0.9333 - val_mae_euclidean: 20.3219 - val_jacard_coef: 0.0345
Epoch 33/200
8/8 [==============================] - 5s 609ms/step - loss: 0.4023 - mae_euclidean: 1.1486 - jacard_coef: 0.4272 - val_loss: 0.9007 - val_mae_euclidean: 11.5399 - val_jacard_coef: 0.0524
Epoch 34/200
8/8 [==============================] - 5s 607ms/step - loss: 0.3916 - mae_euclidean: 1.1199 - jacard_coef: 0.4383 - val_loss: 0.9158 - val_mae_euclidean: 15.8605 - val_jacard_coef: 0.0440
Epoch 35/200
8/8 [==============================] - 5s 607ms/step - loss: 0.4012 - mae_euclidean: 1.6328 - jacard_coef: 0.4306 - val_loss: 0.9354 - val_mae_euclidean: 15.8822 - val_jacard_coef: 0.0334
Epoch 36/200
8/8 [==============================] - 5s 606ms/step - loss: 0.3704 - mae_euclidean: 1.0919 - jacard_coef: 0.4604 - val_loss: 0.9263 - val_mae_euclidean: 18.4639 - val_jacard_coef: 0.0383
Epoch 37/200
8/8 [==============================] - 5s 608ms/step - loss: 0.3603 - mae_euclidean: 1.0440 - jacard_coef: 0.4708 - val_loss: 0.8796 - val_mae_euclidean: 10.0463 - val_jacard_coef: 0.0641
Epoch 38/200
8/8 [==============================] - 5s 608ms/step - loss: 0.3699 - mae_euclidean: 1.1810 - jacard_coef: 0.4641 - val_loss: 0.9018 - val_mae_euclidean: 11.5337 - val_jacard_coef: 0.0516
Epoch 39/200
8/8 [==============================] - 5s 608ms/step - loss: 0.3455 - mae_euclidean: 0.9330 - jacard_coef: 0.4878 - val_loss: 0.8898 - val_mae_euclidean: 10.9618 - val_jacard_coef: 0.0583
Epoch 40/200
8/8 [==============================] - 5s 609ms/step - loss: 0.3367 - mae_euclidean: 0.9277 - jacard_coef: 0.4967 - val_loss: 0.8282 - val_mae_euclidean: 7.1753 - val_jacard_coef: 0.0942
Epoch 41/200
8/8 [==============================] - 5s 607ms/step - loss: 0.3415 - mae_euclidean: 0.9575 - jacard_coef: 0.4916 - val_loss: 0.8777 - val_mae_euclidean: 8.5606 - val_jacard_coef: 0.0652
Epoch 42/200
8/8 [==============================] - 5s 609ms/step - loss: 0.3422 - mae_euclidean: 1.1084 - jacard_coef: 0.4911 - val_loss: 0.8497 - val_mae_euclidean: 8.1112 - val_jacard_coef: 0.0813
Epoch 43/200
8/8 [==============================] - 5s 609ms/step - loss: 0.3271 - mae_euclidean: 0.9384 - jacard_coef: 0.5079 - val_loss: 0.8187 - val_mae_euclidean: 7.2131 - val_jacard_coef: 0.0998
Epoch 44/200
8/8 [==============================] - 5s 609ms/step - loss: 0.3279 - mae_euclidean: 1.0506 - jacard_coef: 0.5073 - val_loss: 0.7975 - val_mae_euclidean: 6.3796 - val_jacard_coef: 0.1129
Epoch 45/200
8/8 [==============================] - 5s 608ms/step - loss: 0.3303 - mae_euclidean: 0.9820 - jacard_coef: 0.5056 - val_loss: 0.8020 - val_mae_euclidean: 6.1947 - val_jacard_coef: 0.1102
Epoch 46/200
8/8 [==============================] - 5s 610ms/step - loss: 0.3312 - mae_euclidean: 1.0105 - jacard_coef: 0.5044 - val_loss: 0.7128 - val_mae_euclidean: 5.7995 - val_jacard_coef: 0.1678
Epoch 47/200
8/8 [==============================] - 5s 608ms/step - loss: 0.3162 - mae_euclidean: 0.9692 - jacard_coef: 0.5204 - val_loss: 0.7246 - val_mae_euclidean: 5.7926 - val_jacard_coef: 0.1599
Epoch 48/200
8/8 [==============================] - 5s 610ms/step - loss: 0.3025 - mae_euclidean: 0.9358 - jacard_coef: 0.5362 - val_loss: 0.6993 - val_mae_euclidean: 5.3684 - val_jacard_coef: 0.1779
Epoch 49/200
8/8 [==============================] - 5s 611ms/step - loss: 0.2893 - mae_euclidean: 0.8107 - jacard_coef: 0.5517 - val_loss: 0.7214 - val_mae_euclidean: 4.9237 - val_jacard_coef: 0.1625
Epoch 50/200
8/8 [==============================] - 5s 614ms/step - loss: 0.2921 - mae_euclidean: 0.8298 - jacard_coef: 0.5485 - val_loss: 0.7943 - val_mae_euclidean: 6.2637 - val_jacard_coef: 0.1150
Epoch 51/200
8/8 [==============================] - 5s 611ms/step - loss: 0.3021 - mae_euclidean: 0.9762 - jacard_coef: 0.5364 - val_loss: 0.6669 - val_mae_euclidean: 4.9943 - val_jacard_coef: 0.2002
Epoch 52/200
8/8 [==============================] - 5s 612ms/step - loss: 0.2966 - mae_euclidean: 0.8238 - jacard_coef: 0.5432 - val_loss: 0.6788 - val_mae_euclidean: 4.6027 - val_jacard_coef: 0.1926
Epoch 53/200
8/8 [==============================] - 5s 612ms/step - loss: 0.2825 - mae_euclidean: 0.7762 - jacard_coef: 0.5598 - val_loss: 0.6395 - val_mae_euclidean: 3.8339 - val_jacard_coef: 0.2218
Epoch 54/200
8/8 [==============================] - 5s 612ms/step - loss: 0.2741 - mae_euclidean: 0.7314 - jacard_coef: 0.5700 - val_loss: 0.5702 - val_mae_euclidean: 3.8825 - val_jacard_coef: 0.2748
Epoch 55/200
8/8 [==============================] - 5s 614ms/step - loss: 0.2766 - mae_euclidean: 0.8394 - jacard_coef: 0.5685 - val_loss: 0.6874 - val_mae_euclidean: 5.1167 - val_jacard_coef: 0.1859
Epoch 56/200
8/8 [==============================] - 5s 610ms/step - loss: 0.2881 - mae_euclidean: 0.8652 - jacard_coef: 0.5556 - val_loss: 0.5607 - val_mae_euclidean: 3.5050 - val_jacard_coef: 0.2833
Epoch 57/200
8/8 [==============================] - 5s 610ms/step - loss: 0.2602 - mae_euclidean: 0.8202 - jacard_coef: 0.5875 - val_loss: 0.5213 - val_mae_euclidean: 3.2169 - val_jacard_coef: 0.3185
Epoch 58/200
8/8 [==============================] - 5s 610ms/step - loss: 0.2549 - mae_euclidean: 0.8654 - jacard_coef: 0.5941 - val_loss: 0.5373 - val_mae_euclidean: 3.2261 - val_jacard_coef: 0.3044
Epoch 59/200
8/8 [==============================] - 5s 608ms/step - loss: 0.2489 - mae_euclidean: 0.8006 - jacard_coef: 0.6017 - val_loss: 0.6149 - val_mae_euclidean: 3.8476 - val_jacard_coef: 0.2397
Epoch 60/200
8/8 [==============================] - 5s 610ms/step - loss: 0.2652 - mae_euclidean: 1.0125 - jacard_coef: 0.5819 - val_loss: 0.4978 - val_mae_euclidean: 3.0929 - val_jacard_coef: 0.3366
Epoch 61/200
8/8 [==============================] - 5s 610ms/step - loss: 0.2516 - mae_euclidean: 0.7029 - jacard_coef: 0.5981 - val_loss: 0.5004 - val_mae_euclidean: 3.1981 - val_jacard_coef: 0.3374
Epoch 62/200
8/8 [==============================] - 5s 608ms/step - loss: 0.2396 - mae_euclidean: 0.7570 - jacard_coef: 0.6141 - val_loss: 0.5128 - val_mae_euclidean: 3.3195 - val_jacard_coef: 0.3231
Epoch 63/200
8/8 [==============================] - 5s 609ms/step - loss: 0.2755 - mae_euclidean: 0.7792 - jacard_coef: 0.5692 - val_loss: 0.4893 - val_mae_euclidean: 3.1399 - val_jacard_coef: 0.3451
Epoch 64/200
8/8 [==============================] - 5s 612ms/step - loss: 0.2848 - mae_euclidean: 0.9411 - jacard_coef: 0.5589 - val_loss: 0.5018 - val_mae_euclidean: 3.1243 - val_jacard_coef: 0.3337
Epoch 65/200
8/8 [==============================] - 5s 612ms/step - loss: 0.2541 - mae_euclidean: 0.9402 - jacard_coef: 0.5969 - val_loss: 0.4383 - val_mae_euclidean: 2.8092 - val_jacard_coef: 0.3927
Epoch 66/200
8/8 [==============================] - 5s 612ms/step - loss: 0.2667 - mae_euclidean: 0.8846 - jacard_coef: 0.5806 - val_loss: 0.4016 - val_mae_euclidean: 2.0892 - val_jacard_coef: 0.4298
Epoch 67/200
8/8 [==============================] - 5s 611ms/step - loss: 0.2512 - mae_euclidean: 0.7424 - jacard_coef: 0.5990 - val_loss: 0.4008 - val_mae_euclidean: 2.1738 - val_jacard_coef: 0.4296
Epoch 68/200
8/8 [==============================] - 5s 612ms/step - loss: 0.2395 - mae_euclidean: 0.7713 - jacard_coef: 0.6144 - val_loss: 0.3528 - val_mae_euclidean: 1.4928 - val_jacard_coef: 0.4808
Epoch 69/200
8/8 [==============================] - 5s 611ms/step - loss: 0.2369 - mae_euclidean: 0.7496 - jacard_coef: 0.6177 - val_loss: 0.3529 - val_mae_euclidean: 1.3945 - val_jacard_coef: 0.4803
Epoch 70/200
8/8 [==============================] - 5s 609ms/step - loss: 0.2213 - mae_euclidean: 0.7163 - jacard_coef: 0.6380 - val_loss: 0.4293 - val_mae_euclidean: 2.6484 - val_jacard_coef: 0.4009
Epoch 71/200
8/8 [==============================] - 5s 607ms/step - loss: 0.2433 - mae_euclidean: 0.8074 - jacard_coef: 0.6104 - val_loss: 0.3815 - val_mae_euclidean: 2.1519 - val_jacard_coef: 0.4492
Epoch 72/200
8/8 [==============================] - 5s 609ms/step - loss: 0.2133 - mae_euclidean: 0.6500 - jacard_coef: 0.6487 - val_loss: 0.3675 - val_mae_euclidean: 1.6960 - val_jacard_coef: 0.4638
Epoch 73/200
8/8 [==============================] - 5s 609ms/step - loss: 0.2170 - mae_euclidean: 0.6736 - jacard_coef: 0.6439 - val_loss: 0.3878 - val_mae_euclidean: 1.3345 - val_jacard_coef: 0.4446
Epoch 74/200
8/8 [==============================] - 5s 614ms/step - loss: 0.2298 - mae_euclidean: 0.6974 - jacard_coef: 0.6278 - val_loss: 0.3614 - val_mae_euclidean: 1.5674 - val_jacard_coef: 0.4700
Epoch 75/200
8/8 [==============================] - 5s 614ms/step - loss: 0.2180 - mae_euclidean: 0.6765 - jacard_coef: 0.6429 - val_loss: 0.3565 - val_mae_euclidean: 1.3579 - val_jacard_coef: 0.4754
Epoch 76/200
8/8 [==============================] - 5s 611ms/step - loss: 0.2275 - mae_euclidean: 0.6522 - jacard_coef: 0.6305 - val_loss: 0.3503 - val_mae_euclidean: 1.1337 - val_jacard_coef: 0.4840
Epoch 77/200
8/8 [==============================] - 5s 614ms/step - loss: 0.2257 - mae_euclidean: 0.8305 - jacard_coef: 0.6336 - val_loss: 0.3945 - val_mae_euclidean: 1.4522 - val_jacard_coef: 0.4351
Epoch 78/200
8/8 [==============================] - 5s 614ms/step - loss: 0.2034 - mae_euclidean: 0.5747 - jacard_coef: 0.6624 - val_loss: 0.3643 - val_mae_euclidean: 1.6962 - val_jacard_coef: 0.4667
Epoch 79/200
8/8 [==============================] - 5s 610ms/step - loss: 0.1871 - mae_euclidean: 0.5508 - jacard_coef: 0.6855 - val_loss: 0.3742 - val_mae_euclidean: 1.9555 - val_jacard_coef: 0.4568
Epoch 80/200
8/8 [==============================] - 5s 611ms/step - loss: 0.1838 - mae_euclidean: 0.5627 - jacard_coef: 0.6904 - val_loss: 0.3814 - val_mae_euclidean: 1.9120 - val_jacard_coef: 0.4488
Epoch 81/200
8/8 [==============================] - 5s 607ms/step - loss: 0.1805 - mae_euclidean: 0.5079 - jacard_coef: 0.6950 - val_loss: 0.3629 - val_mae_euclidean: 1.6425 - val_jacard_coef: 0.4687
Epoch 82/200
8/8 [==============================] - 5s 609ms/step - loss: 0.1739 - mae_euclidean: 0.5638 - jacard_coef: 0.7045 - val_loss: 0.3749 - val_mae_euclidean: 1.5853 - val_jacard_coef: 0.4554
Epoch 83/200
8/8 [==============================] - 5s 607ms/step - loss: 0.1579 - mae_euclidean: 0.4554 - jacard_coef: 0.7274 - val_loss: 0.3806 - val_mae_euclidean: 1.5563 - val_jacard_coef: 0.4495
Epoch 84/200
8/8 [==============================] - 5s 608ms/step - loss: 0.1583 - mae_euclidean: 0.4432 - jacard_coef: 0.7271 - val_loss: 0.3628 - val_mae_euclidean: 1.3823 - val_jacard_coef: 0.4686
Epoch 85/200
8/8 [==============================] - 5s 610ms/step - loss: 0.1730 - mae_euclidean: 0.5104 - jacard_coef: 0.7056 - val_loss: 0.3454 - val_mae_euclidean: 1.1774 - val_jacard_coef: 0.4876
Epoch 86/200
8/8 [==============================] - 5s 607ms/step - loss: 0.1663 - mae_euclidean: 0.5925 - jacard_coef: 0.7153 - val_loss: 0.3511 - val_mae_euclidean: 1.3100 - val_jacard_coef: 0.4815
Epoch 87/200
8/8 [==============================] - 5s 607ms/step - loss: 0.1559 - mae_euclidean: 0.4611 - jacard_coef: 0.7313 - val_loss: 0.3890 - val_mae_euclidean: 1.4781 - val_jacard_coef: 0.4405
Epoch 88/200
8/8 [==============================] - 5s 606ms/step - loss: 0.1524 - mae_euclidean: 0.4709 - jacard_coef: 0.7356 - val_loss: 0.3519 - val_mae_euclidean: 1.2116 - val_jacard_coef: 0.4803
Epoch 89/200
8/8 [==============================] - 5s 608ms/step - loss: 0.1446 - mae_euclidean: 0.4077 - jacard_coef: 0.7477 - val_loss: 0.3719 - val_mae_euclidean: 1.2638 - val_jacard_coef: 0.4594
Epoch 90/200
8/8 [==============================] - 5s 607ms/step - loss: 0.1389 - mae_euclidean: 0.3582 - jacard_coef: 0.7563 - val_loss: 0.3564 - val_mae_euclidean: 1.1883 - val_jacard_coef: 0.4752
Epoch 91/200
8/8 [==============================] - 5s 608ms/step - loss: 0.1249 - mae_euclidean: 0.3778 - jacard_coef: 0.7783 - val_loss: 0.3552 - val_mae_euclidean: 1.2749 - val_jacard_coef: 0.4772
Epoch 92/200
8/8 [==============================] - 5s 606ms/step - loss: 0.1335 - mae_euclidean: 0.4555 - jacard_coef: 0.7650 - val_loss: 0.3507 - val_mae_euclidean: 1.2442 - val_jacard_coef: 0.4819
Epoch 93/200
8/8 [==============================] - 5s 609ms/step - loss: 0.1497 - mae_euclidean: 0.4695 - jacard_coef: 0.7410 - val_loss: 0.3444 - val_mae_euclidean: 1.1984 - val_jacard_coef: 0.4887
Epoch 94/200
8/8 [==============================] - 5s 606ms/step - loss: 0.1288 - mae_euclidean: 0.4333 - jacard_coef: 0.7721 - val_loss: 0.3549 - val_mae_euclidean: 1.1833 - val_jacard_coef: 0.4775
Epoch 95/200
8/8 [==============================] - 5s 607ms/step - loss: 0.1353 - mae_euclidean: 0.3887 - jacard_coef: 0.7622 - val_loss: 0.3653 - val_mae_euclidean: 1.3282 - val_jacard_coef: 0.4654
Epoch 96/200
8/8 [==============================] - 5s 611ms/step - loss: 0.1318 - mae_euclidean: 0.4792 - jacard_coef: 0.7676 - val_loss: 0.3388 - val_mae_euclidean: 1.1051 - val_jacard_coef: 0.4949
Epoch 97/200
8/8 [==============================] - 5s 611ms/step - loss: 0.1168 - mae_euclidean: 0.3406 - jacard_coef: 0.7910 - val_loss: 0.3398 - val_mae_euclidean: 1.2240 - val_jacard_coef: 0.4939
Epoch 98/200
8/8 [==============================] - 5s 611ms/step - loss: 0.1104 - mae_euclidean: 0.3391 - jacard_coef: 0.8014 - val_loss: 0.3479 - val_mae_euclidean: 1.1362 - val_jacard_coef: 0.4848
Epoch 99/200
8/8 [==============================] - 5s 613ms/step - loss: 0.1123 - mae_euclidean: 0.2849 - jacard_coef: 0.7984 - val_loss: 0.3441 - val_mae_euclidean: 1.0627 - val_jacard_coef: 0.4894
Epoch 100/200
8/8 [==============================] - 5s 617ms/step - loss: 0.1022 - mae_euclidean: 0.3117 - jacard_coef: 0.8148 - val_loss: 0.3405 - val_mae_euclidean: 1.1958 - val_jacard_coef: 0.4936
Epoch 101/200
8/8 [==============================] - 5s 616ms/step - loss: 0.0954 - mae_euclidean: 0.3330 - jacard_coef: 0.8261 - val_loss: 0.3541 - val_mae_euclidean: 0.9266 - val_jacard_coef: 0.4788
Epoch 102/200
8/8 [==============================] - 5s 617ms/step - loss: 0.0918 - mae_euclidean: 0.3050 - jacard_coef: 0.8320 - val_loss: 0.3516 - val_mae_euclidean: 1.1770 - val_jacard_coef: 0.4806
Epoch 103/200
8/8 [==============================] - 5s 615ms/step - loss: 0.0981 - mae_euclidean: 0.2992 - jacard_coef: 0.8220 - val_loss: 0.3535 - val_mae_euclidean: 1.1093 - val_jacard_coef: 0.4791
Epoch 104/200
8/8 [==============================] - 5s 615ms/step - loss: 0.1126 - mae_euclidean: 0.2683 - jacard_coef: 0.7987 - val_loss: 0.3457 - val_mae_euclidean: 1.1714 - val_jacard_coef: 0.4875
Epoch 105/200
8/8 [==============================] - 5s 613ms/step - loss: 0.1004 - mae_euclidean: 0.3487 - jacard_coef: 0.8185 - val_loss: 0.3579 - val_mae_euclidean: 1.1319 - val_jacard_coef: 0.4737
Epoch 106/200
8/8 [==============================] - 5s 615ms/step - loss: 0.1228 - mae_euclidean: 0.5543 - jacard_coef: 0.7822 - val_loss: 0.3720 - val_mae_euclidean: 1.3449 - val_jacard_coef: 0.4583
Epoch 107/200
8/8 [==============================] - 5s 613ms/step - loss: 0.1117 - mae_euclidean: 0.3522 - jacard_coef: 0.7993 - val_loss: 0.3556 - val_mae_euclidean: 1.1752 - val_jacard_coef: 0.4771
Epoch 108/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0983 - mae_euclidean: 0.2943 - jacard_coef: 0.8212 - val_loss: 0.3546 - val_mae_euclidean: 1.0912 - val_jacard_coef: 0.4782
Epoch 109/200
8/8 [==============================] - 5s 611ms/step - loss: 0.1011 - mae_euclidean: 0.3997 - jacard_coef: 0.8169 - val_loss: 0.3642 - val_mae_euclidean: 1.0987 - val_jacard_coef: 0.4670
Epoch 110/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0946 - mae_euclidean: 0.2842 - jacard_coef: 0.8277 - val_loss: 0.3403 - val_mae_euclidean: 1.1789 - val_jacard_coef: 0.4937
Epoch 111/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0798 - mae_euclidean: 0.2790 - jacard_coef: 0.8525 - val_loss: 0.3416 - val_mae_euclidean: 1.0845 - val_jacard_coef: 0.4924
Epoch 112/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0806 - mae_euclidean: 0.2435 - jacard_coef: 0.8516 - val_loss: 0.3555 - val_mae_euclidean: 1.2185 - val_jacard_coef: 0.4764
Epoch 113/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0787 - mae_euclidean: 0.2251 - jacard_coef: 0.8542 - val_loss: 0.3521 - val_mae_euclidean: 1.0745 - val_jacard_coef: 0.4799
Epoch 114/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0807 - mae_euclidean: 0.2702 - jacard_coef: 0.8509 - val_loss: 0.3586 - val_mae_euclidean: 1.2088 - val_jacard_coef: 0.4727
Epoch 115/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0701 - mae_euclidean: 0.2311 - jacard_coef: 0.8692 - val_loss: 0.3532 - val_mae_euclidean: 1.0219 - val_jacard_coef: 0.4789
Epoch 116/200
8/8 [==============================] - 5s 608ms/step - loss: 0.0663 - mae_euclidean: 0.2212 - jacard_coef: 0.8759 - val_loss: 0.3498 - val_mae_euclidean: 1.1275 - val_jacard_coef: 0.4824
Epoch 117/200
8/8 [==============================] - 5s 608ms/step - loss: 0.0687 - mae_euclidean: 0.2214 - jacard_coef: 0.8716 - val_loss: 0.3527 - val_mae_euclidean: 0.9479 - val_jacard_coef: 0.4797
Epoch 118/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0628 - mae_euclidean: 0.2101 - jacard_coef: 0.8820 - val_loss: 0.3505 - val_mae_euclidean: 0.9437 - val_jacard_coef: 0.4818
Epoch 119/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0647 - mae_euclidean: 0.1846 - jacard_coef: 0.8788 - val_loss: 0.3517 - val_mae_euclidean: 1.0621 - val_jacard_coef: 0.4811
Epoch 120/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0582 - mae_euclidean: 0.2892 - jacard_coef: 0.8901 - val_loss: 0.3434 - val_mae_euclidean: 1.0731 - val_jacard_coef: 0.4902
Epoch 121/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0607 - mae_euclidean: 0.2088 - jacard_coef: 0.8859 - val_loss: 0.3579 - val_mae_euclidean: 0.9268 - val_jacard_coef: 0.4742
Epoch 122/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0521 - mae_euclidean: 0.1639 - jacard_coef: 0.9011 - val_loss: 0.3495 - val_mae_euclidean: 0.9374 - val_jacard_coef: 0.4832
Epoch 123/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0542 - mae_euclidean: 0.1920 - jacard_coef: 0.8977 - val_loss: 0.3573 - val_mae_euclidean: 1.0320 - val_jacard_coef: 0.4749
Epoch 124/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0504 - mae_euclidean: 0.2032 - jacard_coef: 0.9042 - val_loss: 0.3450 - val_mae_euclidean: 1.0372 - val_jacard_coef: 0.4884
Epoch 125/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0505 - mae_euclidean: 0.1748 - jacard_coef: 0.9039 - val_loss: 0.3523 - val_mae_euclidean: 1.1676 - val_jacard_coef: 0.4806
Epoch 126/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0465 - mae_euclidean: 0.1763 - jacard_coef: 0.9113 - val_loss: 0.3408 - val_mae_euclidean: 0.8436 - val_jacard_coef: 0.4929
Epoch 127/200
8/8 [==============================] - 5s 614ms/step - loss: 0.0455 - mae_euclidean: 0.1743 - jacard_coef: 0.9130 - val_loss: 0.3517 - val_mae_euclidean: 1.2290 - val_jacard_coef: 0.4812
Epoch 128/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0432 - mae_euclidean: 0.1383 - jacard_coef: 0.9172 - val_loss: 0.3472 - val_mae_euclidean: 1.0750 - val_jacard_coef: 0.4861
Epoch 129/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0419 - mae_euclidean: 0.1315 - jacard_coef: 0.9198 - val_loss: 0.3579 - val_mae_euclidean: 1.0498 - val_jacard_coef: 0.4742
Epoch 130/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0419 - mae_euclidean: 0.1490 - jacard_coef: 0.9196 - val_loss: 0.3600 - val_mae_euclidean: 1.2206 - val_jacard_coef: 0.4717
Epoch 131/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0420 - mae_euclidean: 0.1740 - jacard_coef: 0.9195 - val_loss: 0.3464 - val_mae_euclidean: 0.9100 - val_jacard_coef: 0.4870
Epoch 132/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0409 - mae_euclidean: 0.1316 - jacard_coef: 0.9215 - val_loss: 0.3485 - val_mae_euclidean: 1.0660 - val_jacard_coef: 0.4848
Epoch 133/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0393 - mae_euclidean: 0.1261 - jacard_coef: 0.9245 - val_loss: 0.3496 - val_mae_euclidean: 1.0285 - val_jacard_coef: 0.4833
Epoch 134/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0403 - mae_euclidean: 0.1273 - jacard_coef: 0.9227 - val_loss: 0.3541 - val_mae_euclidean: 1.0455 - val_jacard_coef: 0.4790
Epoch 135/200
8/8 [==============================] - 5s 615ms/step - loss: 0.0388 - mae_euclidean: 0.1275 - jacard_coef: 0.9253 - val_loss: 0.3497 - val_mae_euclidean: 1.0033 - val_jacard_coef: 0.4834
Epoch 136/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0368 - mae_euclidean: 0.1586 - jacard_coef: 0.9290 - val_loss: 0.3454 - val_mae_euclidean: 0.9854 - val_jacard_coef: 0.4879
Epoch 137/200
8/8 [==============================] - 5s 617ms/step - loss: 0.0360 - mae_euclidean: 0.1179 - jacard_coef: 0.9307 - val_loss: 0.3541 - val_mae_euclidean: 0.9457 - val_jacard_coef: 0.4785
Epoch 138/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0367 - mae_euclidean: 0.1178 - jacard_coef: 0.9292 - val_loss: 0.3480 - val_mae_euclidean: 0.9147 - val_jacard_coef: 0.4852
Epoch 139/200
8/8 [==============================] - 5s 616ms/step - loss: 0.0332 - mae_euclidean: 0.1262 - jacard_coef: 0.9358 - val_loss: 0.3487 - val_mae_euclidean: 0.9969 - val_jacard_coef: 0.4843
Epoch 140/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0314 - mae_euclidean: 0.1159 - jacard_coef: 0.9393 - val_loss: 0.3451 - val_mae_euclidean: 1.0745 - val_jacard_coef: 0.4884
Epoch 141/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0322 - mae_euclidean: 0.1261 - jacard_coef: 0.9378 - val_loss: 0.3496 - val_mae_euclidean: 1.0258 - val_jacard_coef: 0.4835
Epoch 142/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0312 - mae_euclidean: 0.1128 - jacard_coef: 0.9396 - val_loss: 0.3457 - val_mae_euclidean: 1.0330 - val_jacard_coef: 0.4879
Epoch 143/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0298 - mae_euclidean: 0.1101 - jacard_coef: 0.9421 - val_loss: 0.3529 - val_mae_euclidean: 1.0199 - val_jacard_coef: 0.4798
Epoch 144/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0291 - mae_euclidean: 0.1123 - jacard_coef: 0.9435 - val_loss: 0.3482 - val_mae_euclidean: 1.0565 - val_jacard_coef: 0.4850
Epoch 145/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0288 - mae_euclidean: 0.1094 - jacard_coef: 0.9441 - val_loss: 0.3517 - val_mae_euclidean: 0.9107 - val_jacard_coef: 0.4812
Epoch 146/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0287 - mae_euclidean: 0.1089 - jacard_coef: 0.9443 - val_loss: 0.3469 - val_mae_euclidean: 1.0526 - val_jacard_coef: 0.4866
Epoch 147/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0281 - mae_euclidean: 0.1060 - jacard_coef: 0.9454 - val_loss: 0.3513 - val_mae_euclidean: 0.9456 - val_jacard_coef: 0.4815
Epoch 148/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0295 - mae_euclidean: 0.1263 - jacard_coef: 0.9428 - val_loss: 0.3457 - val_mae_euclidean: 0.9993 - val_jacard_coef: 0.4876
Epoch 149/200
8/8 [==============================] - 5s 616ms/step - loss: 0.0284 - mae_euclidean: 0.1150 - jacard_coef: 0.9448 - val_loss: 0.3484 - val_mae_euclidean: 0.9251 - val_jacard_coef: 0.4849
Epoch 150/200
8/8 [==============================] - 5s 616ms/step - loss: 0.0299 - mae_euclidean: 0.1059 - jacard_coef: 0.9421 - val_loss: 0.3494 - val_mae_euclidean: 0.8727 - val_jacard_coef: 0.4834
Epoch 151/200
8/8 [==============================] - 5s 616ms/step - loss: 0.0273 - mae_euclidean: 0.1053 - jacard_coef: 0.9470 - val_loss: 0.3454 - val_mae_euclidean: 0.8657 - val_jacard_coef: 0.4882
Epoch 152/200
8/8 [==============================] - 5s 617ms/step - loss: 0.0282 - mae_euclidean: 0.1054 - jacard_coef: 0.9452 - val_loss: 0.3497 - val_mae_euclidean: 0.9167 - val_jacard_coef: 0.4832
Epoch 153/200
8/8 [==============================] - 5s 614ms/step - loss: 0.0265 - mae_euclidean: 0.1047 - jacard_coef: 0.9485 - val_loss: 0.3491 - val_mae_euclidean: 0.9298 - val_jacard_coef: 0.4838
Epoch 154/200
8/8 [==============================] - 5s 614ms/step - loss: 0.0269 - mae_euclidean: 0.1033 - jacard_coef: 0.9476 - val_loss: 0.3488 - val_mae_euclidean: 0.9507 - val_jacard_coef: 0.4844
Epoch 155/200
8/8 [==============================] - 5s 616ms/step - loss: 0.0278 - mae_euclidean: 0.1045 - jacard_coef: 0.9460 - val_loss: 0.3483 - val_mae_euclidean: 0.9504 - val_jacard_coef: 0.4848
Epoch 156/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0273 - mae_euclidean: 0.1037 - jacard_coef: 0.9469 - val_loss: 0.3491 - val_mae_euclidean: 0.9390 - val_jacard_coef: 0.4841
Epoch 157/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0265 - mae_euclidean: 0.1031 - jacard_coef: 0.9484 - val_loss: 0.3467 - val_mae_euclidean: 1.0608 - val_jacard_coef: 0.4864
Epoch 158/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0280 - mae_euclidean: 0.1048 - jacard_coef: 0.9456 - val_loss: 0.3515 - val_mae_euclidean: 0.9522 - val_jacard_coef: 0.4814
Epoch 159/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0272 - mae_euclidean: 0.1049 - jacard_coef: 0.9471 - val_loss: 0.3456 - val_mae_euclidean: 0.9382 - val_jacard_coef: 0.4877
Epoch 160/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0256 - mae_euclidean: 0.1023 - jacard_coef: 0.9502 - val_loss: 0.3469 - val_mae_euclidean: 0.8817 - val_jacard_coef: 0.4866
Epoch 161/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0256 - mae_euclidean: 0.1017 - jacard_coef: 0.9502 - val_loss: 0.3457 - val_mae_euclidean: 0.8592 - val_jacard_coef: 0.4879
Epoch 162/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0252 - mae_euclidean: 0.1016 - jacard_coef: 0.9508 - val_loss: 0.3463 - val_mae_euclidean: 0.9182 - val_jacard_coef: 0.4870
Epoch 163/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0252 - mae_euclidean: 0.1007 - jacard_coef: 0.9508 - val_loss: 0.3467 - val_mae_euclidean: 0.9461 - val_jacard_coef: 0.4866
Epoch 164/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0251 - mae_euclidean: 0.1016 - jacard_coef: 0.9511 - val_loss: 0.3499 - val_mae_euclidean: 0.8847 - val_jacard_coef: 0.4834
Epoch 165/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0249 - mae_euclidean: 0.1012 - jacard_coef: 0.9513 - val_loss: 0.3468 - val_mae_euclidean: 0.9378 - val_jacard_coef: 0.4871
Epoch 166/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0259 - mae_euclidean: 0.1015 - jacard_coef: 0.9496 - val_loss: 0.3452 - val_mae_euclidean: 0.8768 - val_jacard_coef: 0.4884
Epoch 167/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0249 - mae_euclidean: 0.1012 - jacard_coef: 0.9514 - val_loss: 0.3467 - val_mae_euclidean: 0.9784 - val_jacard_coef: 0.4867
Epoch 168/200
8/8 [==============================] - 5s 608ms/step - loss: 0.0255 - mae_euclidean: 0.1001 - jacard_coef: 0.9505 - val_loss: 0.3452 - val_mae_euclidean: 0.8669 - val_jacard_coef: 0.4884
Epoch 169/200
8/8 [==============================] - 5s 608ms/step - loss: 0.0245 - mae_euclidean: 0.0995 - jacard_coef: 0.9522 - val_loss: 0.3438 - val_mae_euclidean: 0.8604 - val_jacard_coef: 0.4900
Epoch 170/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0242 - mae_euclidean: 0.1000 - jacard_coef: 0.9528 - val_loss: 0.3457 - val_mae_euclidean: 0.8262 - val_jacard_coef: 0.4878
Epoch 171/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0256 - mae_euclidean: 0.1043 - jacard_coef: 0.9502 - val_loss: 0.3455 - val_mae_euclidean: 0.8391 - val_jacard_coef: 0.4878
Epoch 172/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0264 - mae_euclidean: 0.1001 - jacard_coef: 0.9486 - val_loss: 0.3494 - val_mae_euclidean: 0.9405 - val_jacard_coef: 0.4840
Epoch 173/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0246 - mae_euclidean: 0.1021 - jacard_coef: 0.9521 - val_loss: 0.3484 - val_mae_euclidean: 0.8648 - val_jacard_coef: 0.4850
Epoch 174/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0247 - mae_euclidean: 0.1018 - jacard_coef: 0.9519 - val_loss: 0.3463 - val_mae_euclidean: 0.8529 - val_jacard_coef: 0.4872
Epoch 175/200
8/8 [==============================] - 5s 614ms/step - loss: 0.0251 - mae_euclidean: 0.0997 - jacard_coef: 0.9511 - val_loss: 0.3448 - val_mae_euclidean: 1.0493 - val_jacard_coef: 0.4887
Epoch 176/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0261 - mae_euclidean: 0.0997 - jacard_coef: 0.9492 - val_loss: 0.3434 - val_mae_euclidean: 0.8713 - val_jacard_coef: 0.4900
Epoch 177/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0243 - mae_euclidean: 0.0984 - jacard_coef: 0.9525 - val_loss: 0.3477 - val_mae_euclidean: 0.8527 - val_jacard_coef: 0.4857
Epoch 178/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0244 - mae_euclidean: 0.0978 - jacard_coef: 0.9524 - val_loss: 0.3457 - val_mae_euclidean: 0.9851 - val_jacard_coef: 0.4876
Epoch 179/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0242 - mae_euclidean: 0.0977 - jacard_coef: 0.9527 - val_loss: 0.3493 - val_mae_euclidean: 0.8550 - val_jacard_coef: 0.4838
Epoch 180/200
8/8 [==============================] - 5s 610ms/step - loss: 0.0249 - mae_euclidean: 0.0978 - jacard_coef: 0.9515 - val_loss: 0.3449 - val_mae_euclidean: 1.0683 - val_jacard_coef: 0.4887
Epoch 181/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0244 - mae_euclidean: 0.0976 - jacard_coef: 0.9525 - val_loss: 0.3439 - val_mae_euclidean: 0.9658 - val_jacard_coef: 0.4896
Epoch 182/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0240 - mae_euclidean: 0.0973 - jacard_coef: 0.9532 - val_loss: 0.3440 - val_mae_euclidean: 0.8300 - val_jacard_coef: 0.4896
Epoch 183/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0241 - mae_euclidean: 0.0978 - jacard_coef: 0.9530 - val_loss: 0.3467 - val_mae_euclidean: 0.9822 - val_jacard_coef: 0.4864
Epoch 184/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0242 - mae_euclidean: 0.0975 - jacard_coef: 0.9529 - val_loss: 0.3499 - val_mae_euclidean: 0.8985 - val_jacard_coef: 0.4830
Epoch 185/200
8/8 [==============================] - 5s 609ms/step - loss: 0.0232 - mae_euclidean: 0.0990 - jacard_coef: 0.9546 - val_loss: 0.3456 - val_mae_euclidean: 0.8932 - val_jacard_coef: 0.4878
Epoch 186/200
8/8 [==============================] - 5s 607ms/step - loss: 0.0226 - mae_euclidean: 0.0965 - jacard_coef: 0.9558 - val_loss: 0.3439 - val_mae_euclidean: 0.8962 - val_jacard_coef: 0.4898
Epoch 187/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0225 - mae_euclidean: 0.0963 - jacard_coef: 0.9560 - val_loss: 0.3425 - val_mae_euclidean: 0.9130 - val_jacard_coef: 0.4913
Epoch 188/200
8/8 [==============================] - 5s 612ms/step - loss: 0.0222 - mae_euclidean: 0.0974 - jacard_coef: 0.9566 - val_loss: 0.3404 - val_mae_euclidean: 0.9079 - val_jacard_coef: 0.4936
Epoch 189/200
8/8 [==============================] - 5s 613ms/step - loss: 0.0225 - mae_euclidean: 0.0955 - jacard_coef: 0.9560 - val_loss: 0.3400 - val_mae_euclidean: 0.8637 - val_jacard_coef: 0.4940
Epoch 190/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0219 - mae_euclidean: 0.0950 - jacard_coef: 0.9571 - val_loss: 0.3414 - val_mae_euclidean: 0.9499 - val_jacard_coef: 0.4925
Epoch 191/200
8/8 [==============================] - 5s 608ms/step - loss: 0.0218 - mae_euclidean: 0.0949 - jacard_coef: 0.9573 - val_loss: 0.3415 - val_mae_euclidean: 0.9004 - val_jacard_coef: 0.4923
Epoch 192/200
8/8 [==============================] - 5s 607ms/step - loss: 0.0219 - mae_euclidean: 0.0944 - jacard_coef: 0.9571 - val_loss: 0.3411 - val_mae_euclidean: 0.8714 - val_jacard_coef: 0.4929
Epoch 193/200
8/8 [==============================] - 5s 606ms/step - loss: 0.0215 - mae_euclidean: 0.0943 - jacard_coef: 0.9579 - val_loss: 0.3416 - val_mae_euclidean: 0.8866 - val_jacard_coef: 0.4922
Epoch 194/200
8/8 [==============================] - 5s 607ms/step - loss: 0.0219 - mae_euclidean: 0.0943 - jacard_coef: 0.9571 - val_loss: 0.3423 - val_mae_euclidean: 1.0056 - val_jacard_coef: 0.4915
Epoch 195/200
8/8 [==============================] - 5s 608ms/step - loss: 0.0232 - mae_euclidean: 0.0951 - jacard_coef: 0.9547 - val_loss: 0.3422 - val_mae_euclidean: 1.0017 - val_jacard_coef: 0.4916
Epoch 196/200
8/8 [==============================] - 5s 607ms/step - loss: 0.0226 - mae_euclidean: 0.0947 - jacard_coef: 0.9559 - val_loss: 0.3420 - val_mae_euclidean: 0.8921 - val_jacard_coef: 0.4918
Epoch 197/200
8/8 [==============================] - 5s 608ms/step - loss: 0.0223 - mae_euclidean: 0.0940 - jacard_coef: 0.9565 - val_loss: 0.3426 - val_mae_euclidean: 0.9359 - val_jacard_coef: 0.4911
Epoch 198/200
8/8 [==============================] - 5s 604ms/step - loss: 0.0225 - mae_euclidean: 0.0943 - jacard_coef: 0.9560 - val_loss: 0.3419 - val_mae_euclidean: 0.9045 - val_jacard_coef: 0.4918
Epoch 199/200
8/8 [==============================] - 5s 611ms/step - loss: 0.0223 - mae_euclidean: 0.0941 - jacard_coef: 0.9564 - val_loss: 0.3412 - val_mae_euclidean: 0.9230 - val_jacard_coef: 0.4926
Epoch 200/200
8/8 [==============================] - 5s 608ms/step - loss: 0.0213 - mae_euclidean: 0.0945 - jacard_coef: 0.9583 - val_loss: 0.3427 - val_mae_euclidean: 0.9038 - val_jacard_coef: 0.4908
Model trained for 975.937887430191s

plot_history(history_unet1, "Unet")

model_unet1.evaluate(X_val, y_val)

1/1 [==============================] - 1s 977ms/step - loss: 0.3261 - mae_euclidean: 0.9038 - jacard_coef: 0.5082

[0.3261321187019348, 0.9037566781044006, 0.5081525444984436]

Attention residual UNet model trials

tf.keras.backend.clear_session()

# Attention residual Unet1
tf.random.set_seed(14)
model_att_res_unet1 = Attention_ResUNet(input_shape)
model_att_res_unet1.compile(optimizer=Adam(learning_rate=1e-2),
                     loss = dice_coef_loss,
                     metrics=[mae_euclidean, jacard_coef]
                     )
start = time.time()
history_att_res_unet1 = model_att_res_unet1.fit(X_train, y_train,
                                validation_data = (X_val, y_val),
                                batch_size = 2,
                                epochs=100
                               )
print(f"Model trained for {time.time() - start}s")
model_att_res_unet1.save("2022-02-17 att_res_UNet1 100epochs.hdf5")

Epoch 1/100
16/16 [==============================] - 21s 465ms/step - loss: 0.8110 - mae_euclidean: 3.5976 - jacard_coef: 0.1057 - val_loss: 0.8614 - val_mae_euclidean: 6.3043 - val_jacard_coef: 0.0756
Epoch 2/100
16/16 [==============================] - 7s 419ms/step - loss: 0.7632 - mae_euclidean: 3.8068 - jacard_coef: 0.1357 - val_loss: 0.9898 - val_mae_euclidean: 120.9765 - val_jacard_coef: 0.0052
Epoch 3/100
16/16 [==============================] - 7s 417ms/step - loss: 0.7324 - mae_euclidean: 3.6451 - jacard_coef: 0.1557 - val_loss: 0.9974 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 0.0014
Epoch 4/100
16/16 [==============================] - 7s 418ms/step - loss: 0.7036 - mae_euclidean: 3.0018 - jacard_coef: 0.1753 - val_loss: 0.8520 - val_mae_euclidean: 6.2725 - val_jacard_coef: 0.0811
Epoch 5/100
16/16 [==============================] - 7s 418ms/step - loss: 0.6764 - mae_euclidean: 2.7265 - jacard_coef: 0.1947 - val_loss: 0.8727 - val_mae_euclidean: 41.8583 - val_jacard_coef: 0.0686
Epoch 6/100
16/16 [==============================] - 7s 419ms/step - loss: 0.6512 - mae_euclidean: 2.7744 - jacard_coef: 0.2145 - val_loss: 0.8903 - val_mae_euclidean: 40.4907 - val_jacard_coef: 0.0583
Epoch 7/100
16/16 [==============================] - 7s 419ms/step - loss: 0.6222 - mae_euclidean: 2.3448 - jacard_coef: 0.2364 - val_loss: 0.9147 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0448
Epoch 8/100
16/16 [==============================] - 7s 418ms/step - loss: 0.5890 - mae_euclidean: 2.1212 - jacard_coef: 0.2611 - val_loss: 0.9205 - val_mae_euclidean: 98.6868 - val_jacard_coef: 0.0416
Epoch 9/100
16/16 [==============================] - 7s 432ms/step - loss: 0.5765 - mae_euclidean: 2.1199 - jacard_coef: 0.2725 - val_loss: 0.9285 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 0.0372
Epoch 10/100
16/16 [==============================] - 8s 473ms/step - loss: 0.5334 - mae_euclidean: 1.8830 - jacard_coef: 0.3067 - val_loss: 0.9321 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 0.0352
Epoch 11/100
16/16 [==============================] - 8s 474ms/step - loss: 0.5203 - mae_euclidean: 1.9309 - jacard_coef: 0.3199 - val_loss: 0.9132 - val_mae_euclidean: 23.2299 - val_jacard_coef: 0.0456
Epoch 12/100
16/16 [==============================] - 8s 472ms/step - loss: 0.4959 - mae_euclidean: 1.8279 - jacard_coef: 0.3414 - val_loss: 0.8894 - val_mae_euclidean: 12.8182 - val_jacard_coef: 0.0589
Epoch 13/100
16/16 [==============================] - 8s 473ms/step - loss: 0.4647 - mae_euclidean: 1.3266 - jacard_coef: 0.3691 - val_loss: 0.9066 - val_mae_euclidean: 15.2139 - val_jacard_coef: 0.0493
Epoch 14/100
16/16 [==============================] - 8s 472ms/step - loss: 0.4450 - mae_euclidean: 1.2176 - jacard_coef: 0.3875 - val_loss: 0.8838 - val_mae_euclidean: 11.1062 - val_jacard_coef: 0.0623
Epoch 15/100
16/16 [==============================] - 8s 471ms/step - loss: 0.4182 - mae_euclidean: 1.1773 - jacard_coef: 0.4127 - val_loss: 0.9148 - val_mae_euclidean: 14.7130 - val_jacard_coef: 0.0448
Epoch 16/100
16/16 [==============================] - 8s 471ms/step - loss: 0.4000 - mae_euclidean: 0.9952 - jacard_coef: 0.4307 - val_loss: 0.7669 - val_mae_euclidean: 5.7636 - val_jacard_coef: 0.1343
Epoch 17/100
16/16 [==============================] - 8s 473ms/step - loss: 0.3946 - mae_euclidean: 1.0846 - jacard_coef: 0.4365 - val_loss: 0.8551 - val_mae_euclidean: 10.4260 - val_jacard_coef: 0.0792
Epoch 18/100
16/16 [==============================] - 8s 473ms/step - loss: 0.3859 - mae_euclidean: 1.0698 - jacard_coef: 0.4451 - val_loss: 0.8763 - val_mae_euclidean: 11.7109 - val_jacard_coef: 0.0666
Epoch 19/100
16/16 [==============================] - 8s 473ms/step - loss: 0.3790 - mae_euclidean: 1.0971 - jacard_coef: 0.4532 - val_loss: 0.6938 - val_mae_euclidean: 4.8333 - val_jacard_coef: 0.1835
Epoch 20/100
16/16 [==============================] - 8s 475ms/step - loss: 0.3669 - mae_euclidean: 1.1033 - jacard_coef: 0.4653 - val_loss: 0.8790 - val_mae_euclidean: 10.9072 - val_jacard_coef: 0.0650
Epoch 21/100
16/16 [==============================] - 8s 474ms/step - loss: 0.3524 - mae_euclidean: 0.9955 - jacard_coef: 0.4814 - val_loss: 0.7020 - val_mae_euclidean: 6.2855 - val_jacard_coef: 0.1768
Epoch 22/100
16/16 [==============================] - 8s 476ms/step - loss: 0.3274 - mae_euclidean: 0.8990 - jacard_coef: 0.5093 - val_loss: 0.7164 - val_mae_euclidean: 6.1164 - val_jacard_coef: 0.1679
Epoch 23/100
16/16 [==============================] - 8s 476ms/step - loss: 0.3191 - mae_euclidean: 0.8380 - jacard_coef: 0.5186 - val_loss: 0.6612 - val_mae_euclidean: 4.4105 - val_jacard_coef: 0.2067
Epoch 24/100
16/16 [==============================] - 8s 475ms/step - loss: 0.3023 - mae_euclidean: 0.7076 - jacard_coef: 0.5375 - val_loss: 0.7645 - val_mae_euclidean: 6.9083 - val_jacard_coef: 0.1354
Epoch 25/100
16/16 [==============================] - 8s 475ms/step - loss: 0.2877 - mae_euclidean: 0.7486 - jacard_coef: 0.5547 - val_loss: 0.5409 - val_mae_euclidean: 3.3326 - val_jacard_coef: 0.3002
Epoch 26/100
16/16 [==============================] - 8s 476ms/step - loss: 0.2816 - mae_euclidean: 0.6883 - jacard_coef: 0.5621 - val_loss: 0.7035 - val_mae_euclidean: 6.4583 - val_jacard_coef: 0.1762
Epoch 27/100
16/16 [==============================] - 8s 478ms/step - loss: 0.2690 - mae_euclidean: 0.6014 - jacard_coef: 0.5777 - val_loss: 0.6178 - val_mae_euclidean: 4.1309 - val_jacard_coef: 0.2402
Epoch 28/100
16/16 [==============================] - 8s 478ms/step - loss: 0.2637 - mae_euclidean: 0.7349 - jacard_coef: 0.5849 - val_loss: 0.6690 - val_mae_euclidean: 5.9285 - val_jacard_coef: 0.2004
Epoch 29/100
16/16 [==============================] - 8s 477ms/step - loss: 0.2643 - mae_euclidean: 0.7235 - jacard_coef: 0.5837 - val_loss: 0.5395 - val_mae_euclidean: 3.3782 - val_jacard_coef: 0.3012
Epoch 30/100
16/16 [==============================] - 8s 477ms/step - loss: 0.2625 - mae_euclidean: 0.5945 - jacard_coef: 0.5856 - val_loss: 0.5121 - val_mae_euclidean: 2.8072 - val_jacard_coef: 0.3248
Epoch 31/100
16/16 [==============================] - 8s 476ms/step - loss: 0.2660 - mae_euclidean: 0.6366 - jacard_coef: 0.5813 - val_loss: 0.5590 - val_mae_euclidean: 3.8379 - val_jacard_coef: 0.2850
Epoch 32/100
16/16 [==============================] - 8s 475ms/step - loss: 0.2416 - mae_euclidean: 0.5648 - jacard_coef: 0.6121 - val_loss: 0.5713 - val_mae_euclidean: 3.8166 - val_jacard_coef: 0.2749
Epoch 33/100
16/16 [==============================] - 8s 475ms/step - loss: 0.2245 - mae_euclidean: 0.4858 - jacard_coef: 0.6341 - val_loss: 0.5495 - val_mae_euclidean: 3.7925 - val_jacard_coef: 0.2921
Epoch 34/100
16/16 [==============================] - 8s 475ms/step - loss: 0.2012 - mae_euclidean: 0.4392 - jacard_coef: 0.6658 - val_loss: 0.4539 - val_mae_euclidean: 2.0257 - val_jacard_coef: 0.3765
Epoch 35/100
16/16 [==============================] - 8s 477ms/step - loss: 0.2106 - mae_euclidean: 0.4870 - jacard_coef: 0.6534 - val_loss: 0.5188 - val_mae_euclidean: 2.3950 - val_jacard_coef: 0.3180
Epoch 36/100
16/16 [==============================] - 8s 477ms/step - loss: 0.2064 - mae_euclidean: 0.4830 - jacard_coef: 0.6594 - val_loss: 0.4520 - val_mae_euclidean: 1.7992 - val_jacard_coef: 0.3784
Epoch 37/100
16/16 [==============================] - 8s 474ms/step - loss: 0.1969 - mae_euclidean: 0.4132 - jacard_coef: 0.6726 - val_loss: 0.4365 - val_mae_euclidean: 1.8925 - val_jacard_coef: 0.3933
Epoch 38/100
16/16 [==============================] - 8s 476ms/step - loss: 0.2154 - mae_euclidean: 0.4769 - jacard_coef: 0.6465 - val_loss: 0.4243 - val_mae_euclidean: 1.3546 - val_jacard_coef: 0.4046
Epoch 39/100
16/16 [==============================] - 8s 475ms/step - loss: 0.1780 - mae_euclidean: 0.3702 - jacard_coef: 0.6984 - val_loss: 0.3866 - val_mae_euclidean: 1.2989 - val_jacard_coef: 0.4434
Epoch 40/100
16/16 [==============================] - 8s 475ms/step - loss: 0.1585 - mae_euclidean: 0.3463 - jacard_coef: 0.7269 - val_loss: 0.3764 - val_mae_euclidean: 1.2643 - val_jacard_coef: 0.4541
Epoch 41/100
16/16 [==============================] - 8s 477ms/step - loss: 0.1556 - mae_euclidean: 0.2970 - jacard_coef: 0.7312 - val_loss: 0.3962 - val_mae_euclidean: 1.3295 - val_jacard_coef: 0.4326
Epoch 42/100
16/16 [==============================] - 8s 478ms/step - loss: 0.1609 - mae_euclidean: 0.3339 - jacard_coef: 0.7234 - val_loss: 0.3798 - val_mae_euclidean: 1.1066 - val_jacard_coef: 0.4504
Epoch 43/100
16/16 [==============================] - 8s 477ms/step - loss: 0.1460 - mae_euclidean: 0.3090 - jacard_coef: 0.7455 - val_loss: 0.3733 - val_mae_euclidean: 1.1689 - val_jacard_coef: 0.4572
Epoch 44/100
16/16 [==============================] - 8s 476ms/step - loss: 0.1542 - mae_euclidean: 0.3197 - jacard_coef: 0.7339 - val_loss: 0.4212 - val_mae_euclidean: 1.2646 - val_jacard_coef: 0.4105
Epoch 45/100
16/16 [==============================] - 8s 478ms/step - loss: 0.1514 - mae_euclidean: 0.2985 - jacard_coef: 0.7378 - val_loss: 0.4044 - val_mae_euclidean: 1.2754 - val_jacard_coef: 0.4258
Epoch 46/100
16/16 [==============================] - 8s 477ms/step - loss: 0.1614 - mae_euclidean: 0.3831 - jacard_coef: 0.7232 - val_loss: 0.3989 - val_mae_euclidean: 1.3110 - val_jacard_coef: 0.4314
Epoch 47/100
16/16 [==============================] - 8s 478ms/step - loss: 0.1466 - mae_euclidean: 0.3098 - jacard_coef: 0.7447 - val_loss: 0.4231 - val_mae_euclidean: 1.4616 - val_jacard_coef: 0.4059
Epoch 48/100
16/16 [==============================] - 8s 478ms/step - loss: 0.1323 - mae_euclidean: 0.2791 - jacard_coef: 0.7670 - val_loss: 0.3774 - val_mae_euclidean: 1.1209 - val_jacard_coef: 0.4525
Epoch 49/100
16/16 [==============================] - 8s 474ms/step - loss: 0.1174 - mae_euclidean: 0.2349 - jacard_coef: 0.7899 - val_loss: 0.4040 - val_mae_euclidean: 1.2483 - val_jacard_coef: 0.4253
Epoch 50/100
16/16 [==============================] - 8s 475ms/step - loss: 0.1109 - mae_euclidean: 0.2168 - jacard_coef: 0.8006 - val_loss: 0.3735 - val_mae_euclidean: 1.2091 - val_jacard_coef: 0.4577
Epoch 51/100
16/16 [==============================] - 8s 472ms/step - loss: 0.1075 - mae_euclidean: 0.2096 - jacard_coef: 0.8061 - val_loss: 0.3698 - val_mae_euclidean: 1.0784 - val_jacard_coef: 0.4613
Epoch 52/100
16/16 [==============================] - 8s 474ms/step - loss: 0.1143 - mae_euclidean: 0.2199 - jacard_coef: 0.7952 - val_loss: 0.3785 - val_mae_euclidean: 0.9192 - val_jacard_coef: 0.4516
Epoch 53/100
16/16 [==============================] - 8s 475ms/step - loss: 0.1095 - mae_euclidean: 0.2508 - jacard_coef: 0.8028 - val_loss: 0.3726 - val_mae_euclidean: 1.0267 - val_jacard_coef: 0.4577
Epoch 54/100
16/16 [==============================] - 8s 475ms/step - loss: 0.1101 - mae_euclidean: 0.2150 - jacard_coef: 0.8022 - val_loss: 0.3752 - val_mae_euclidean: 0.9777 - val_jacard_coef: 0.4555
Epoch 55/100
16/16 [==============================] - 8s 475ms/step - loss: 0.1048 - mae_euclidean: 0.1995 - jacard_coef: 0.8107 - val_loss: 0.3789 - val_mae_euclidean: 1.2099 - val_jacard_coef: 0.4520
Epoch 56/100
16/16 [==============================] - 8s 476ms/step - loss: 0.1250 - mae_euclidean: 0.2594 - jacard_coef: 0.7784 - val_loss: 0.3903 - val_mae_euclidean: 1.1307 - val_jacard_coef: 0.4396
Epoch 57/100
16/16 [==============================] - 8s 477ms/step - loss: 0.1031 - mae_euclidean: 0.2129 - jacard_coef: 0.8135 - val_loss: 0.3643 - val_mae_euclidean: 1.2201 - val_jacard_coef: 0.4664
Epoch 58/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0917 - mae_euclidean: 0.2168 - jacard_coef: 0.8322 - val_loss: 0.3622 - val_mae_euclidean: 1.2110 - val_jacard_coef: 0.4691
Epoch 59/100
16/16 [==============================] - 8s 478ms/step - loss: 0.0853 - mae_euclidean: 0.1790 - jacard_coef: 0.8431 - val_loss: 0.3515 - val_mae_euclidean: 1.1944 - val_jacard_coef: 0.4806
Epoch 60/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0966 - mae_euclidean: 0.1956 - jacard_coef: 0.8242 - val_loss: 0.4730 - val_mae_euclidean: 1.4013 - val_jacard_coef: 0.3588
Epoch 61/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0832 - mae_euclidean: 0.1926 - jacard_coef: 0.8467 - val_loss: 0.3623 - val_mae_euclidean: 1.2515 - val_jacard_coef: 0.4688
Epoch 62/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0794 - mae_euclidean: 0.1601 - jacard_coef: 0.8531 - val_loss: 0.3622 - val_mae_euclidean: 1.0571 - val_jacard_coef: 0.4688
Epoch 63/100
16/16 [==============================] - 8s 474ms/step - loss: 0.0748 - mae_euclidean: 0.1609 - jacard_coef: 0.8610 - val_loss: 0.3611 - val_mae_euclidean: 0.9694 - val_jacard_coef: 0.4701
Epoch 64/100
16/16 [==============================] - 8s 475ms/step - loss: 0.0838 - mae_euclidean: 0.1814 - jacard_coef: 0.8455 - val_loss: 0.3796 - val_mae_euclidean: 1.0853 - val_jacard_coef: 0.4503
Epoch 65/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0798 - mae_euclidean: 0.1697 - jacard_coef: 0.8523 - val_loss: 0.3598 - val_mae_euclidean: 1.1530 - val_jacard_coef: 0.4722
Epoch 66/100
16/16 [==============================] - 8s 478ms/step - loss: 0.0880 - mae_euclidean: 0.2241 - jacard_coef: 0.8383 - val_loss: 0.3674 - val_mae_euclidean: 1.0259 - val_jacard_coef: 0.4633
Epoch 67/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0771 - mae_euclidean: 0.1643 - jacard_coef: 0.8571 - val_loss: 0.3615 - val_mae_euclidean: 1.0411 - val_jacard_coef: 0.4695
Epoch 68/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0705 - mae_euclidean: 0.1486 - jacard_coef: 0.8686 - val_loss: 0.3519 - val_mae_euclidean: 0.9884 - val_jacard_coef: 0.4804
Epoch 69/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0653 - mae_euclidean: 0.1496 - jacard_coef: 0.8776 - val_loss: 0.3562 - val_mae_euclidean: 0.8855 - val_jacard_coef: 0.4754
Epoch 70/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0609 - mae_euclidean: 0.1370 - jacard_coef: 0.8855 - val_loss: 0.3601 - val_mae_euclidean: 1.0064 - val_jacard_coef: 0.4709
Epoch 71/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0601 - mae_euclidean: 0.1353 - jacard_coef: 0.8867 - val_loss: 0.3732 - val_mae_euclidean: 0.9871 - val_jacard_coef: 0.4574
Epoch 72/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0566 - mae_euclidean: 0.1291 - jacard_coef: 0.8931 - val_loss: 0.3588 - val_mae_euclidean: 1.0163 - val_jacard_coef: 0.4726
Epoch 73/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0550 - mae_euclidean: 0.1228 - jacard_coef: 0.8958 - val_loss: 0.3527 - val_mae_euclidean: 1.0132 - val_jacard_coef: 0.4795
Epoch 74/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0544 - mae_euclidean: 0.1280 - jacard_coef: 0.8970 - val_loss: 0.3556 - val_mae_euclidean: 1.0828 - val_jacard_coef: 0.4764
Epoch 75/100
16/16 [==============================] - 8s 478ms/step - loss: 0.0503 - mae_euclidean: 0.1241 - jacard_coef: 0.9045 - val_loss: 0.3596 - val_mae_euclidean: 1.1509 - val_jacard_coef: 0.4716
Epoch 76/100
16/16 [==============================] - 8s 478ms/step - loss: 0.0498 - mae_euclidean: 0.1230 - jacard_coef: 0.9052 - val_loss: 0.3589 - val_mae_euclidean: 1.0735 - val_jacard_coef: 0.4724
Epoch 77/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0479 - mae_euclidean: 0.1162 - jacard_coef: 0.9088 - val_loss: 0.3676 - val_mae_euclidean: 1.0676 - val_jacard_coef: 0.4630
Epoch 78/100
16/16 [==============================] - 8s 475ms/step - loss: 0.0528 - mae_euclidean: 0.1257 - jacard_coef: 0.8999 - val_loss: 0.3587 - val_mae_euclidean: 1.1314 - val_jacard_coef: 0.4728
Epoch 79/100
16/16 [==============================] - 8s 474ms/step - loss: 0.0479 - mae_euclidean: 0.1179 - jacard_coef: 0.9087 - val_loss: 0.3635 - val_mae_euclidean: 1.1336 - val_jacard_coef: 0.4674
Epoch 80/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0457 - mae_euclidean: 0.1141 - jacard_coef: 0.9127 - val_loss: 0.3480 - val_mae_euclidean: 1.0228 - val_jacard_coef: 0.4845
Epoch 81/100
16/16 [==============================] - 8s 475ms/step - loss: 0.0429 - mae_euclidean: 0.1069 - jacard_coef: 0.9179 - val_loss: 0.3548 - val_mae_euclidean: 1.1559 - val_jacard_coef: 0.4772
Epoch 82/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0441 - mae_euclidean: 0.1066 - jacard_coef: 0.9155 - val_loss: 0.3689 - val_mae_euclidean: 1.0511 - val_jacard_coef: 0.4619
Epoch 83/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0414 - mae_euclidean: 0.1047 - jacard_coef: 0.9206 - val_loss: 0.3570 - val_mae_euclidean: 0.9969 - val_jacard_coef: 0.4746
Epoch 84/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0389 - mae_euclidean: 0.1018 - jacard_coef: 0.9252 - val_loss: 0.3520 - val_mae_euclidean: 0.9345 - val_jacard_coef: 0.4806
Epoch 85/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0397 - mae_euclidean: 0.1069 - jacard_coef: 0.9237 - val_loss: 0.3639 - val_mae_euclidean: 0.9792 - val_jacard_coef: 0.4675
Epoch 86/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0376 - mae_euclidean: 0.1040 - jacard_coef: 0.9276 - val_loss: 0.3627 - val_mae_euclidean: 1.0703 - val_jacard_coef: 0.4686
Epoch 87/100
16/16 [==============================] - 8s 477ms/step - loss: 0.0371 - mae_euclidean: 0.1025 - jacard_coef: 0.9287 - val_loss: 0.3534 - val_mae_euclidean: 0.9960 - val_jacard_coef: 0.4791
Epoch 88/100
16/16 [==============================] - 8s 475ms/step - loss: 0.0359 - mae_euclidean: 0.1006 - jacard_coef: 0.9308 - val_loss: 0.3541 - val_mae_euclidean: 1.1114 - val_jacard_coef: 0.4782
Epoch 89/100
16/16 [==============================] - 8s 476ms/step - loss: 0.0350 - mae_euclidean: 0.0973 - jacard_coef: 0.9326 - val_loss: 0.3578 - val_mae_euclidean: 1.0285 - val_jacard_coef: 0.4744
Epoch 90/100
16/16 [==============================] - 8s 480ms/step - loss: 0.0339 - mae_euclidean: 0.0976 - jacard_coef: 0.9345 - val_loss: 0.3535 - val_mae_euclidean: 0.9266 - val_jacard_coef: 0.4786
Epoch 91/100
16/16 [==============================] - 8s 481ms/step - loss: 0.0332 - mae_euclidean: 0.0937 - jacard_coef: 0.9359 - val_loss: 0.3530 - val_mae_euclidean: 0.8607 - val_jacard_coef: 0.4795
Epoch 92/100
16/16 [==============================] - 8s 481ms/step - loss: 0.0323 - mae_euclidean: 0.0890 - jacard_coef: 0.9376 - val_loss: 0.3506 - val_mae_euclidean: 0.9518 - val_jacard_coef: 0.4822
Epoch 93/100
16/16 [==============================] - 8s 480ms/step - loss: 0.0319 - mae_euclidean: 0.0876 - jacard_coef: 0.9382 - val_loss: 0.3511 - val_mae_euclidean: 0.9609 - val_jacard_coef: 0.4815
Epoch 94/100
16/16 [==============================] - 8s 481ms/step - loss: 0.0306 - mae_euclidean: 0.0868 - jacard_coef: 0.9406 - val_loss: 0.3529 - val_mae_euclidean: 0.9655 - val_jacard_coef: 0.4794
Epoch 95/100
16/16 [==============================] - 8s 480ms/step - loss: 0.0309 - mae_euclidean: 0.0849 - jacard_coef: 0.9402 - val_loss: 0.3553 - val_mae_euclidean: 0.9034 - val_jacard_coef: 0.4768
Epoch 96/100
16/16 [==============================] - 8s 481ms/step - loss: 0.0298 - mae_euclidean: 0.0857 - jacard_coef: 0.9423 - val_loss: 0.3594 - val_mae_euclidean: 0.9253 - val_jacard_coef: 0.4724
Epoch 97/100
16/16 [==============================] - 8s 481ms/step - loss: 0.0290 - mae_euclidean: 0.0857 - jacard_coef: 0.9436 - val_loss: 0.3535 - val_mae_euclidean: 0.8919 - val_jacard_coef: 0.4787
Epoch 98/100
16/16 [==============================] - 8s 481ms/step - loss: 0.0289 - mae_euclidean: 0.0855 - jacard_coef: 0.9439 - val_loss: 0.3536 - val_mae_euclidean: 0.8694 - val_jacard_coef: 0.4789
Epoch 99/100
16/16 [==============================] - 8s 486ms/step - loss: 0.0276 - mae_euclidean: 0.1021 - jacard_coef: 0.9463 - val_loss: 0.3550 - val_mae_euclidean: 1.0335 - val_jacard_coef: 0.4770
Epoch 100/100
16/16 [==============================] - 8s 488ms/step - loss: 0.0281 - mae_euclidean: 0.0840 - jacard_coef: 0.9455 - val_loss: 0.3526 - val_mae_euclidean: 1.0255 - val_jacard_coef: 0.4797
Model trained for 764.92422914505s

C:\Users\a1048794\Anaconda3\lib\site-packages\keras\engine\functional.py:1410: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.
  layer_config = serialize_layer_fn(layer)

plot_history(history_att_res_unet1, "Attention residual Unet")

plot_train_val(model_att_res_unet1, 1, 1)

SA Unet tuning - block_size, keep_prob, start_neurons

The results indicate the smallest model, SA UNet gives the best Jacard scores. It is supposed that the small number of channels in the network is the reason for better generalization. Bayesian optimization will be performed.

def model_builder_SA_UNet(hp):
    hp_block_size =  hp.Int('block_size', min_value=15, max_value=35, step=5) # bigger - more regularization
    hp_keep_prob = hp.Float('keep_prob', min_value=0.5, max_value=0.9, step=0.1)
    hp_start_neurons = hp.Int('start_neurons', min_value=10, max_value=30, step=5)
    
    model = SA_UNet(input_shape, block_size=hp_block_size, keep_prob=hp_keep_prob, start_neurons=hp_start_neurons)
    
    model.compile(optimizer = Adam(learning_rate = 1e-2),
                 loss = dice_coef_loss,
                 metrics=[mae_euclidean, jacard_coef])

    return model

tuner = kt.BayesianOptimization(model_builder_SA_UNet,
                                 objective = kt.Objective("val_mae_euclidean", direction='min'), 
                                 max_trials=16,
                                 seed = 14,
                                 directory=os.path.normpath('C:/keras_tuner'),
                                 project_name='2022-02-16_SA_UNet_tune1')

INFO:tensorflow:Reloading Oracle from existing project C:\keras_tuner\2022-02-16_SA_UNet_tune1\oracle.json
INFO:tensorflow:Reloading Tuner from C:\keras_tuner\2022-02-16_SA_UNet_tune1\tuner0.json

# stop_early = tf.keras.callbacks.EarlyStopping(monitor='val_jacard_coef',
#                                               mode='max',
#                                               patience=30)

tf.random.set_seed(14)
tuner.search(X_train, y_train,
             validation_data = (X_val, y_val),
             #callbacks=[stop_early],
             batch_size = 8,
             epochs=200
            ) 

INFO:tensorflow:Oracle triggered exit

tuner.results_summary(num_trials=5)

Results summary
Results in C:\keras_tuner\2022-02-16_SA_UNet_tune1
Showing 5 best trials
Objective(name='val_mae_euclidean', direction='min')
Trial summary
Hyperparameters:
block_size: 25
keep_prob: 0.7999999999999999
start_neurons: 20
Score: 0.8768714666366577
Trial summary
Hyperparameters:
block_size: 35
keep_prob: 0.8999999999999999
start_neurons: 30
Score: 0.9034335613250732
Trial summary
Hyperparameters:
block_size: 20
keep_prob: 0.8999999999999999
start_neurons: 30
Score: 0.9521717429161072
Trial summary
Hyperparameters:
block_size: 35
keep_prob: 0.8999999999999999
start_neurons: 10
Score: 0.9712845087051392
Trial summary
Hyperparameters:
block_size: 30
keep_prob: 0.8999999999999999
start_neurons: 20
Score: 0.97222501039505

Top 5 parameters produce similar scores with wide range of parameters (only keep_prob is always on the upper end). Block size in top 3 models varies between 25, 35, 20 producing objective within the noise.

# SA Unet2
# increasing the block size from 19 to 25 as recomended by the tuner
tf.random.set_seed(14)
model_sa_unet2 = SA_UNet(input_shape, block_size=25, keep_prob=0.8, start_neurons=20)
model_sa_unet2.compile(optimizer=Adam(learning_rate=1e-2),
                         loss = dice_coef_loss,
                         metrics=[mae_euclidean, jacard_coef]
                         )
start = time.time()
history_sa_unet2 = model_sa_unet2.fit(X_train, y_train,
                                    validation_data = (X_val, y_val),
                                    batch_size = 8, # no resources for 16
                                    epochs=200
                                   )
print(f"Model trained for {time.time() - start}s")
model_sa_unet2.save("2022-02-17 SA-UNet2 200epochs.hdf5")

Epoch 1/200
4/4 [==============================] - 8s 565ms/step - loss: 0.8421 - mae_euclidean: 5.2554 - jacard_coef: 0.0865 - val_loss: 0.8597 - val_mae_euclidean: 6.0402 - val_jacard_coef: 0.0755
Epoch 2/200
4/4 [==============================] - 2s 463ms/step - loss: 0.7753 - mae_euclidean: 3.5628 - jacard_coef: 0.1268 - val_loss: 0.8938 - val_mae_euclidean: 5.5043 - val_jacard_coef: 0.0561
Epoch 3/200
4/4 [==============================] - 2s 462ms/step - loss: 0.7370 - mae_euclidean: 3.2331 - jacard_coef: 0.1515 - val_loss: 0.9131 - val_mae_euclidean: 5.3065 - val_jacard_coef: 0.0454
Epoch 4/200
4/4 [==============================] - 2s 464ms/step - loss: 0.6959 - mae_euclidean: 2.9099 - jacard_coef: 0.1794 - val_loss: 0.9897 - val_mae_euclidean: 62.5067 - val_jacard_coef: 0.0052
Epoch 5/200
4/4 [==============================] - 2s 464ms/step - loss: 0.6573 - mae_euclidean: 2.9397 - jacard_coef: 0.2074 - val_loss: 0.8808 - val_mae_euclidean: 5.6331 - val_jacard_coef: 0.0634
Epoch 6/200
4/4 [==============================] - 2s 462ms/step - loss: 0.6138 - mae_euclidean: 2.4209 - jacard_coef: 0.2395 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.2110e-05
Epoch 7/200
4/4 [==============================] - 2s 466ms/step - loss: 0.5826 - mae_euclidean: 2.4039 - jacard_coef: 0.2647 - val_loss: 0.9974 - val_mae_euclidean: 67.6274 - val_jacard_coef: 0.0013
Epoch 8/200
4/4 [==============================] - 2s 468ms/step - loss: 0.5418 - mae_euclidean: 2.1917 - jacard_coef: 0.2974 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.8672e-05
Epoch 9/200
4/4 [==============================] - 2s 463ms/step - loss: 0.5071 - mae_euclidean: 2.0722 - jacard_coef: 0.3274 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.4241e-05
Epoch 10/200
4/4 [==============================] - 2s 463ms/step - loss: 0.4765 - mae_euclidean: 1.6004 - jacard_coef: 0.3546 - val_loss: 0.9161 - val_mae_euclidean: 5.2070 - val_jacard_coef: 0.0438
Epoch 11/200
4/4 [==============================] - 2s 463ms/step - loss: 0.4598 - mae_euclidean: 1.5873 - jacard_coef: 0.3705 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.7682e-05
Epoch 12/200
4/4 [==============================] - 2s 468ms/step - loss: 0.4388 - mae_euclidean: 1.6359 - jacard_coef: 0.3901 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.7969e-05
Epoch 13/200
4/4 [==============================] - 2s 464ms/step - loss: 0.4296 - mae_euclidean: 1.6023 - jacard_coef: 0.3992 - val_loss: 0.9997 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.4880e-04
Epoch 14/200
4/4 [==============================] - 2s 466ms/step - loss: 0.4246 - mae_euclidean: 1.5134 - jacard_coef: 0.4042 - val_loss: 0.9982 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 9.2090e-04
Epoch 15/200
4/4 [==============================] - 2s 462ms/step - loss: 0.4101 - mae_euclidean: 1.4663 - jacard_coef: 0.4184 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.3014e-05
Epoch 16/200
4/4 [==============================] - 2s 466ms/step - loss: 0.4054 - mae_euclidean: 1.6057 - jacard_coef: 0.4232 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.9480e-05
Epoch 17/200
4/4 [==============================] - 2s 466ms/step - loss: 0.3959 - mae_euclidean: 1.3988 - jacard_coef: 0.4329 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.7320e-05
Epoch 18/200
4/4 [==============================] - 2s 464ms/step - loss: 0.4023 - mae_euclidean: 1.4818 - jacard_coef: 0.4263 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.8215e-05
Epoch 19/200
4/4 [==============================] - 2s 468ms/step - loss: 0.3904 - mae_euclidean: 1.4536 - jacard_coef: 0.4385 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.9233e-05
Epoch 20/200
4/4 [==============================] - 2s 468ms/step - loss: 0.3888 - mae_euclidean: 1.4948 - jacard_coef: 0.4402 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.0692e-05
Epoch 21/200
4/4 [==============================] - 2s 467ms/step - loss: 0.3809 - mae_euclidean: 1.4691 - jacard_coef: 0.4486 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.0520e-05
Epoch 22/200
4/4 [==============================] - 2s 468ms/step - loss: 0.3741 - mae_euclidean: 1.4521 - jacard_coef: 0.4555 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0759e-05
Epoch 23/200
4/4 [==============================] - 2s 465ms/step - loss: 0.3817 - mae_euclidean: 1.3111 - jacard_coef: 0.4479 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.8421e-05
Epoch 24/200
4/4 [==============================] - 2s 468ms/step - loss: 0.3637 - mae_euclidean: 1.2886 - jacard_coef: 0.4668 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.0411e-05
Epoch 25/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3635 - mae_euclidean: 1.2626 - jacard_coef: 0.4669 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0675e-05
Epoch 26/200
4/4 [==============================] - 2s 466ms/step - loss: 0.3540 - mae_euclidean: 1.1819 - jacard_coef: 0.4772 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.1880e-05
Epoch 27/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3697 - mae_euclidean: 1.3256 - jacard_coef: 0.4607 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.5565e-05
Epoch 28/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3660 - mae_euclidean: 1.6460 - jacard_coef: 0.4652 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.4278e-05
Epoch 29/200
4/4 [==============================] - 2s 467ms/step - loss: 0.3513 - mae_euclidean: 1.1183 - jacard_coef: 0.4801 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.2345e-05
Epoch 30/200
4/4 [==============================] - 2s 460ms/step - loss: 0.3512 - mae_euclidean: 1.1970 - jacard_coef: 0.4803 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.5283e-05
Epoch 31/200
4/4 [==============================] - 2s 468ms/step - loss: 0.3471 - mae_euclidean: 1.1714 - jacard_coef: 0.4848 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.2340e-05
Epoch 32/200
4/4 [==============================] - 2s 461ms/step - loss: 0.3482 - mae_euclidean: 1.2472 - jacard_coef: 0.4837 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.1170e-05
Epoch 33/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3440 - mae_euclidean: 1.3189 - jacard_coef: 0.4884 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.3057e-05
Epoch 34/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3440 - mae_euclidean: 1.1571 - jacard_coef: 0.4887 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.3485e-05
Epoch 35/200
4/4 [==============================] - 2s 467ms/step - loss: 0.3569 - mae_euclidean: 1.5092 - jacard_coef: 0.4758 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.8228e-05
Epoch 36/200
4/4 [==============================] - 2s 465ms/step - loss: 0.3342 - mae_euclidean: 1.1372 - jacard_coef: 0.4991 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.1091e-05
Epoch 37/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3284 - mae_euclidean: 1.0869 - jacard_coef: 0.5056 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0780e-05
Epoch 38/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3284 - mae_euclidean: 1.1075 - jacard_coef: 0.5058 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.1751e-05
Epoch 39/200
4/4 [==============================] - 2s 470ms/step - loss: 0.3216 - mae_euclidean: 1.1532 - jacard_coef: 0.5133 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.4959e-05
Epoch 40/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3224 - mae_euclidean: 1.2902 - jacard_coef: 0.5125 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0660e-05
Epoch 41/200
4/4 [==============================] - 2s 460ms/step - loss: 0.3214 - mae_euclidean: 1.0837 - jacard_coef: 0.5138 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.9514e-05
Epoch 42/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3226 - mae_euclidean: 1.2557 - jacard_coef: 0.5123 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.2355e-05
Epoch 43/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3144 - mae_euclidean: 1.0794 - jacard_coef: 0.5217 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.5395e-05
Epoch 44/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3166 - mae_euclidean: 1.1171 - jacard_coef: 0.5195 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.7375e-05
Epoch 45/200
4/4 [==============================] - 2s 468ms/step - loss: 0.3140 - mae_euclidean: 1.1272 - jacard_coef: 0.5222 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.1898e-04
Epoch 46/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3325 - mae_euclidean: 1.1674 - jacard_coef: 0.5013 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.7737e-05
Epoch 47/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3205 - mae_euclidean: 1.1249 - jacard_coef: 0.5147 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.7185e-05
Epoch 48/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3145 - mae_euclidean: 1.0473 - jacard_coef: 0.5218 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.1352e-04
Epoch 49/200
4/4 [==============================] - 2s 467ms/step - loss: 0.3192 - mae_euclidean: 1.0638 - jacard_coef: 0.5162 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.3179e-05
Epoch 50/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3075 - mae_euclidean: 1.0588 - jacard_coef: 0.5299 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.2672e-04
Epoch 51/200
4/4 [==============================] - 2s 460ms/step - loss: 0.3135 - mae_euclidean: 1.0540 - jacard_coef: 0.5228 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 9.4840e-05
Epoch 52/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3026 - mae_euclidean: 0.9484 - jacard_coef: 0.5354 - val_loss: 0.9997 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.4827e-04
Epoch 53/200
4/4 [==============================] - 2s 461ms/step - loss: 0.3052 - mae_euclidean: 1.0470 - jacard_coef: 0.5323 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.7977e-05
Epoch 54/200
4/4 [==============================] - 2s 468ms/step - loss: 0.3030 - mae_euclidean: 1.0817 - jacard_coef: 0.5350 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.2459e-05
Epoch 55/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3019 - mae_euclidean: 0.9330 - jacard_coef: 0.5364 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.8881e-05
Epoch 56/200
4/4 [==============================] - 2s 462ms/step - loss: 0.3030 - mae_euclidean: 1.0946 - jacard_coef: 0.5350 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.4050e-04
Epoch 57/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3037 - mae_euclidean: 1.1249 - jacard_coef: 0.5341 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.0412e-04
Epoch 58/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3018 - mae_euclidean: 1.0745 - jacard_coef: 0.5365 - val_loss: 0.9997 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.5027e-04
Epoch 59/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3038 - mae_euclidean: 1.0821 - jacard_coef: 0.5341 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.5653e-05
Epoch 60/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3101 - mae_euclidean: 1.0127 - jacard_coef: 0.5268 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.8375e-05
Epoch 61/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2979 - mae_euclidean: 1.0571 - jacard_coef: 0.5410 - val_loss: 0.9998 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 1.4184e-04
Epoch 62/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3060 - mae_euclidean: 0.9863 - jacard_coef: 0.5315 - val_loss: 0.9996 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 2.3917e-04
Epoch 63/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3124 - mae_euclidean: 1.0006 - jacard_coef: 0.5241 - val_loss: 0.9998 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 1.4123e-04
Epoch 64/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3312 - mae_euclidean: 1.1581 - jacard_coef: 0.5030 - val_loss: 0.9987 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 6.8523e-04
Epoch 65/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3166 - mae_euclidean: 1.0481 - jacard_coef: 0.5192 - val_loss: 0.9954 - val_mae_euclidean: 53.5558 - val_jacard_coef: 0.0023
Epoch 66/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3252 - mae_euclidean: 1.2137 - jacard_coef: 0.5094 - val_loss: 0.9976 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0012
Epoch 67/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3082 - mae_euclidean: 1.0712 - jacard_coef: 0.5288 - val_loss: 0.9980 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0010
Epoch 68/200
4/4 [==============================] - 2s 464ms/step - loss: 0.3039 - mae_euclidean: 1.0123 - jacard_coef: 0.5340 - val_loss: 0.9977 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0012
Epoch 69/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2999 - mae_euclidean: 1.0624 - jacard_coef: 0.5388 - val_loss: 0.9959 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0021
Epoch 70/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2907 - mae_euclidean: 0.9332 - jacard_coef: 0.5496 - val_loss: 0.9967 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0017
Epoch 71/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2969 - mae_euclidean: 1.0269 - jacard_coef: 0.5426 - val_loss: 0.9990 - val_mae_euclidean: 6917528477885267968.0000 - val_jacard_coef: 4.9950e-04
Epoch 72/200
4/4 [==============================] - 2s 466ms/step - loss: 0.2947 - mae_euclidean: 1.0955 - jacard_coef: 0.5449 - val_loss: 0.9960 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0020
Epoch 73/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2886 - mae_euclidean: 1.0048 - jacard_coef: 0.5521 - val_loss: 0.9965 - val_mae_euclidean: 75.6255 - val_jacard_coef: 0.0018
Epoch 74/200
4/4 [==============================] - 2s 462ms/step - loss: 0.2956 - mae_euclidean: 1.1388 - jacard_coef: 0.5440 - val_loss: 0.9922 - val_mae_euclidean: 40.5433 - val_jacard_coef: 0.0040
Epoch 75/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2870 - mae_euclidean: 0.9634 - jacard_coef: 0.5541 - val_loss: 0.9926 - val_mae_euclidean: 51.5498 - val_jacard_coef: 0.0037
Epoch 76/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2862 - mae_euclidean: 0.9544 - jacard_coef: 0.5551 - val_loss: 0.9918 - val_mae_euclidean: 40.8082 - val_jacard_coef: 0.0041
Epoch 77/200
4/4 [==============================] - 2s 463ms/step - loss: 0.3021 - mae_euclidean: 1.3926 - jacard_coef: 0.5371 - val_loss: 0.9842 - val_mae_euclidean: 36.0560 - val_jacard_coef: 0.0080
Epoch 78/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2994 - mae_euclidean: 1.0757 - jacard_coef: 0.5392 - val_loss: 0.9969 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0016
Epoch 79/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2879 - mae_euclidean: 0.9651 - jacard_coef: 0.5531 - val_loss: 0.9957 - val_mae_euclidean: 51.9813 - val_jacard_coef: 0.0022
Epoch 80/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2880 - mae_euclidean: 0.9894 - jacard_coef: 0.5532 - val_loss: 0.9929 - val_mae_euclidean: 45.0087 - val_jacard_coef: 0.0036
Epoch 81/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2896 - mae_euclidean: 0.9247 - jacard_coef: 0.5511 - val_loss: 0.9968 - val_mae_euclidean: 56.5793 - val_jacard_coef: 0.0016
Epoch 82/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2798 - mae_euclidean: 1.0042 - jacard_coef: 0.5629 - val_loss: 0.9886 - val_mae_euclidean: 38.4394 - val_jacard_coef: 0.0058
Epoch 83/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2810 - mae_euclidean: 0.9206 - jacard_coef: 0.5614 - val_loss: 0.9918 - val_mae_euclidean: 60.0607 - val_jacard_coef: 0.0041
Epoch 84/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2836 - mae_euclidean: 0.8802 - jacard_coef: 0.5584 - val_loss: 0.9747 - val_mae_euclidean: 22.5367 - val_jacard_coef: 0.0128
Epoch 85/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2849 - mae_euclidean: 0.9825 - jacard_coef: 0.5566 - val_loss: 0.9705 - val_mae_euclidean: 23.1788 - val_jacard_coef: 0.0150
Epoch 86/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2837 - mae_euclidean: 1.0044 - jacard_coef: 0.5581 - val_loss: 0.9294 - val_mae_euclidean: 15.2648 - val_jacard_coef: 0.0366
Epoch 87/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2807 - mae_euclidean: 0.9504 - jacard_coef: 0.5618 - val_loss: 0.9484 - val_mae_euclidean: 16.6622 - val_jacard_coef: 0.0265
Epoch 88/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2785 - mae_euclidean: 0.8878 - jacard_coef: 0.5644 - val_loss: 0.9633 - val_mae_euclidean: 21.4981 - val_jacard_coef: 0.0187
Epoch 89/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2878 - mae_euclidean: 1.2572 - jacard_coef: 0.5537 - val_loss: 0.8424 - val_mae_euclidean: 9.4738 - val_jacard_coef: 0.0855
Epoch 90/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2838 - mae_euclidean: 1.0824 - jacard_coef: 0.5580 - val_loss: 0.7896 - val_mae_euclidean: 7.7327 - val_jacard_coef: 0.1176
Epoch 91/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2773 - mae_euclidean: 0.9602 - jacard_coef: 0.5658 - val_loss: 0.8412 - val_mae_euclidean: 9.9946 - val_jacard_coef: 0.0862
Epoch 92/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2738 - mae_euclidean: 1.0554 - jacard_coef: 0.5702 - val_loss: 0.8416 - val_mae_euclidean: 9.5445 - val_jacard_coef: 0.0860
Epoch 93/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2903 - mae_euclidean: 1.0668 - jacard_coef: 0.5502 - val_loss: 0.8601 - val_mae_euclidean: 9.4267 - val_jacard_coef: 0.0752
Epoch 94/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2884 - mae_euclidean: 1.1928 - jacard_coef: 0.5527 - val_loss: 0.7775 - val_mae_euclidean: 7.8594 - val_jacard_coef: 0.1252
Epoch 95/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2986 - mae_euclidean: 1.0856 - jacard_coef: 0.5426 - val_loss: 0.7478 - val_mae_euclidean: 6.8446 - val_jacard_coef: 0.1443
Epoch 96/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2725 - mae_euclidean: 1.0373 - jacard_coef: 0.5718 - val_loss: 0.6812 - val_mae_euclidean: 5.9544 - val_jacard_coef: 0.1897
Epoch 97/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2785 - mae_euclidean: 1.0637 - jacard_coef: 0.5646 - val_loss: 0.7318 - val_mae_euclidean: 6.3802 - val_jacard_coef: 0.1549
Epoch 98/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2749 - mae_euclidean: 1.0731 - jacard_coef: 0.5688 - val_loss: 0.7374 - val_mae_euclidean: 6.1436 - val_jacard_coef: 0.1512
Epoch 99/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2720 - mae_euclidean: 1.0631 - jacard_coef: 0.5723 - val_loss: 0.6205 - val_mae_euclidean: 4.8591 - val_jacard_coef: 0.2342
Epoch 100/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2643 - mae_euclidean: 0.8377 - jacard_coef: 0.5820 - val_loss: 0.6673 - val_mae_euclidean: 5.2511 - val_jacard_coef: 0.1996
Epoch 101/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2737 - mae_euclidean: 1.0931 - jacard_coef: 0.5704 - val_loss: 0.6703 - val_mae_euclidean: 5.4202 - val_jacard_coef: 0.1974
Epoch 102/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2737 - mae_euclidean: 0.9871 - jacard_coef: 0.5703 - val_loss: 0.6769 - val_mae_euclidean: 5.7058 - val_jacard_coef: 0.1927
Epoch 103/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2678 - mae_euclidean: 0.8935 - jacard_coef: 0.5775 - val_loss: 0.6196 - val_mae_euclidean: 4.9030 - val_jacard_coef: 0.2348
Epoch 104/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2673 - mae_euclidean: 0.9595 - jacard_coef: 0.5783 - val_loss: 0.7086 - val_mae_euclidean: 5.7214 - val_jacard_coef: 0.1706
Epoch 105/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2642 - mae_euclidean: 0.9188 - jacard_coef: 0.5822 - val_loss: 0.5900 - val_mae_euclidean: 4.3312 - val_jacard_coef: 0.2579
Epoch 106/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2659 - mae_euclidean: 1.0484 - jacard_coef: 0.5800 - val_loss: 0.5859 - val_mae_euclidean: 4.3147 - val_jacard_coef: 0.2611
Epoch 107/200
4/4 [==============================] - 2s 463ms/step - loss: 0.2643 - mae_euclidean: 0.9650 - jacard_coef: 0.5819 - val_loss: 0.5974 - val_mae_euclidean: 4.0939 - val_jacard_coef: 0.2520
Epoch 108/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2661 - mae_euclidean: 0.9911 - jacard_coef: 0.5797 - val_loss: 0.5716 - val_mae_euclidean: 4.2069 - val_jacard_coef: 0.2726
Epoch 109/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2687 - mae_euclidean: 0.8368 - jacard_coef: 0.5765 - val_loss: 0.5736 - val_mae_euclidean: 4.3976 - val_jacard_coef: 0.2709
Epoch 110/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2625 - mae_euclidean: 0.8152 - jacard_coef: 0.5843 - val_loss: 0.6724 - val_mae_euclidean: 5.5033 - val_jacard_coef: 0.1959
Epoch 111/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2562 - mae_euclidean: 0.9108 - jacard_coef: 0.5922 - val_loss: 0.5860 - val_mae_euclidean: 4.5665 - val_jacard_coef: 0.2610
Epoch 112/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2701 - mae_euclidean: 0.9214 - jacard_coef: 0.5757 - val_loss: 0.6249 - val_mae_euclidean: 4.6217 - val_jacard_coef: 0.2309
Epoch 113/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2632 - mae_euclidean: 1.0201 - jacard_coef: 0.5836 - val_loss: 0.6099 - val_mae_euclidean: 4.4375 - val_jacard_coef: 0.2423
Epoch 114/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2627 - mae_euclidean: 0.9691 - jacard_coef: 0.5841 - val_loss: 0.6012 - val_mae_euclidean: 4.5538 - val_jacard_coef: 0.2491
Epoch 115/200
4/4 [==============================] - 2s 462ms/step - loss: 0.2593 - mae_euclidean: 0.7963 - jacard_coef: 0.5884 - val_loss: 0.6097 - val_mae_euclidean: 4.3995 - val_jacard_coef: 0.2425
Epoch 116/200
4/4 [==============================] - 2s 465ms/step - loss: 0.2572 - mae_euclidean: 0.8343 - jacard_coef: 0.5911 - val_loss: 0.5553 - val_mae_euclidean: 4.2329 - val_jacard_coef: 0.2859
Epoch 117/200
4/4 [==============================] - 2s 464ms/step - loss: 0.2573 - mae_euclidean: 0.9376 - jacard_coef: 0.5910 - val_loss: 0.5808 - val_mae_euclidean: 4.5407 - val_jacard_coef: 0.2652
Epoch 118/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2503 - mae_euclidean: 0.8115 - jacard_coef: 0.5997 - val_loss: 0.4867 - val_mae_euclidean: 3.3884 - val_jacard_coef: 0.3452
Epoch 119/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2535 - mae_euclidean: 0.9091 - jacard_coef: 0.5956 - val_loss: 0.5432 - val_mae_euclidean: 4.1913 - val_jacard_coef: 0.2960
Epoch 120/200
4/4 [==============================] - 2s 493ms/step - loss: 0.2477 - mae_euclidean: 0.8881 - jacard_coef: 0.6031 - val_loss: 0.5000 - val_mae_euclidean: 3.5178 - val_jacard_coef: 0.3334
Epoch 121/200
4/4 [==============================] - 2s 492ms/step - loss: 0.2488 - mae_euclidean: 0.8178 - jacard_coef: 0.6016 - val_loss: 0.5268 - val_mae_euclidean: 3.7795 - val_jacard_coef: 0.3099
Epoch 122/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2527 - mae_euclidean: 0.8659 - jacard_coef: 0.5969 - val_loss: 0.4844 - val_mae_euclidean: 3.3347 - val_jacard_coef: 0.3473
Epoch 123/200
4/4 [==============================] - 2s 493ms/step - loss: 0.2618 - mae_euclidean: 0.9829 - jacard_coef: 0.5858 - val_loss: 0.5132 - val_mae_euclidean: 3.2820 - val_jacard_coef: 0.3217
Epoch 124/200
4/4 [==============================] - 2s 494ms/step - loss: 0.2496 - mae_euclidean: 0.8742 - jacard_coef: 0.6006 - val_loss: 0.4293 - val_mae_euclidean: 2.8648 - val_jacard_coef: 0.3993
Epoch 125/200
4/4 [==============================] - 2s 492ms/step - loss: 0.2503 - mae_euclidean: 0.9130 - jacard_coef: 0.6001 - val_loss: 0.4437 - val_mae_euclidean: 2.9306 - val_jacard_coef: 0.3853
Epoch 126/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2639 - mae_euclidean: 0.9117 - jacard_coef: 0.5827 - val_loss: 0.4745 - val_mae_euclidean: 2.9748 - val_jacard_coef: 0.3564
Epoch 127/200
4/4 [==============================] - 2s 491ms/step - loss: 0.2545 - mae_euclidean: 0.9523 - jacard_coef: 0.5943 - val_loss: 0.4262 - val_mae_euclidean: 2.8947 - val_jacard_coef: 0.4023
Epoch 128/200
4/4 [==============================] - 2s 491ms/step - loss: 0.2547 - mae_euclidean: 0.9562 - jacard_coef: 0.5941 - val_loss: 0.3900 - val_mae_euclidean: 2.8866 - val_jacard_coef: 0.4389
Epoch 129/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2517 - mae_euclidean: 0.9122 - jacard_coef: 0.5983 - val_loss: 0.4941 - val_mae_euclidean: 3.9090 - val_jacard_coef: 0.3386
Epoch 130/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2463 - mae_euclidean: 0.8212 - jacard_coef: 0.6049 - val_loss: 0.4677 - val_mae_euclidean: 3.2145 - val_jacard_coef: 0.3627
Epoch 131/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2461 - mae_euclidean: 0.9014 - jacard_coef: 0.6051 - val_loss: 0.4433 - val_mae_euclidean: 3.0033 - val_jacard_coef: 0.3857
Epoch 132/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2606 - mae_euclidean: 0.9452 - jacard_coef: 0.5870 - val_loss: 0.4518 - val_mae_euclidean: 3.0594 - val_jacard_coef: 0.3776
Epoch 133/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2434 - mae_euclidean: 0.9227 - jacard_coef: 0.6086 - val_loss: 0.3760 - val_mae_euclidean: 2.5775 - val_jacard_coef: 0.4535
Epoch 134/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2403 - mae_euclidean: 0.8312 - jacard_coef: 0.6127 - val_loss: 0.4469 - val_mae_euclidean: 2.6990 - val_jacard_coef: 0.3823
Epoch 135/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2437 - mae_euclidean: 0.8282 - jacard_coef: 0.6084 - val_loss: 0.3768 - val_mae_euclidean: 1.8702 - val_jacard_coef: 0.4526
Epoch 136/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2400 - mae_euclidean: 0.9424 - jacard_coef: 0.6130 - val_loss: 0.3766 - val_mae_euclidean: 2.0961 - val_jacard_coef: 0.4529
Epoch 137/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2415 - mae_euclidean: 0.8148 - jacard_coef: 0.6113 - val_loss: 0.4037 - val_mae_euclidean: 2.5450 - val_jacard_coef: 0.4248
Epoch 138/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2451 - mae_euclidean: 1.0723 - jacard_coef: 0.6068 - val_loss: 0.3705 - val_mae_euclidean: 1.3696 - val_jacard_coef: 0.4593
Epoch 139/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2360 - mae_euclidean: 0.7500 - jacard_coef: 0.6182 - val_loss: 0.3770 - val_mae_euclidean: 2.1402 - val_jacard_coef: 0.4525
Epoch 140/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2322 - mae_euclidean: 0.7313 - jacard_coef: 0.6231 - val_loss: 0.3523 - val_mae_euclidean: 1.7587 - val_jacard_coef: 0.4790
Epoch 141/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2425 - mae_euclidean: 0.9713 - jacard_coef: 0.6110 - val_loss: 0.3747 - val_mae_euclidean: 2.4750 - val_jacard_coef: 0.4549
Epoch 142/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2429 - mae_euclidean: 0.8125 - jacard_coef: 0.6092 - val_loss: 0.3493 - val_mae_euclidean: 1.7488 - val_jacard_coef: 0.4823
Epoch 143/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2374 - mae_euclidean: 0.8111 - jacard_coef: 0.6164 - val_loss: 0.3232 - val_mae_euclidean: 1.2571 - val_jacard_coef: 0.5115
Epoch 144/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2376 - mae_euclidean: 0.7906 - jacard_coef: 0.6162 - val_loss: 0.3196 - val_mae_euclidean: 1.4102 - val_jacard_coef: 0.5156
Epoch 145/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2369 - mae_euclidean: 0.9044 - jacard_coef: 0.6171 - val_loss: 0.3076 - val_mae_euclidean: 1.3635 - val_jacard_coef: 0.5295
Epoch 146/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2376 - mae_euclidean: 0.8754 - jacard_coef: 0.6162 - val_loss: 0.3154 - val_mae_euclidean: 1.6977 - val_jacard_coef: 0.5205
Epoch 147/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2346 - mae_euclidean: 0.8266 - jacard_coef: 0.6201 - val_loss: 0.3653 - val_mae_euclidean: 2.4151 - val_jacard_coef: 0.4649
Epoch 148/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2374 - mae_euclidean: 0.8903 - jacard_coef: 0.6164 - val_loss: 0.3165 - val_mae_euclidean: 1.8573 - val_jacard_coef: 0.5192
Epoch 149/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2434 - mae_euclidean: 0.8438 - jacard_coef: 0.6087 - val_loss: 0.3581 - val_mae_euclidean: 1.9169 - val_jacard_coef: 0.4727
Epoch 150/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2435 - mae_euclidean: 1.0658 - jacard_coef: 0.6086 - val_loss: 0.3118 - val_mae_euclidean: 1.1504 - val_jacard_coef: 0.5246
Epoch 151/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2381 - mae_euclidean: 0.7577 - jacard_coef: 0.6154 - val_loss: 0.3222 - val_mae_euclidean: 1.3423 - val_jacard_coef: 0.5126
Epoch 152/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2385 - mae_euclidean: 0.7845 - jacard_coef: 0.6152 - val_loss: 0.3269 - val_mae_euclidean: 1.7041 - val_jacard_coef: 0.5073
Epoch 153/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2362 - mae_euclidean: 0.9409 - jacard_coef: 0.6179 - val_loss: 0.3066 - val_mae_euclidean: 1.5807 - val_jacard_coef: 0.5307
Epoch 154/200
4/4 [==============================] - 2s 491ms/step - loss: 0.2392 - mae_euclidean: 0.8480 - jacard_coef: 0.6141 - val_loss: 0.3382 - val_mae_euclidean: 1.7508 - val_jacard_coef: 0.4945
Epoch 155/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2301 - mae_euclidean: 0.7987 - jacard_coef: 0.6260 - val_loss: 0.3287 - val_mae_euclidean: 1.7576 - val_jacard_coef: 0.5052
Epoch 156/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2286 - mae_euclidean: 0.8420 - jacard_coef: 0.6280 - val_loss: 0.3403 - val_mae_euclidean: 1.3617 - val_jacard_coef: 0.4922
Epoch 157/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2259 - mae_euclidean: 0.7292 - jacard_coef: 0.6317 - val_loss: 0.3199 - val_mae_euclidean: 1.1489 - val_jacard_coef: 0.5153
Epoch 158/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2350 - mae_euclidean: 0.8872 - jacard_coef: 0.6195 - val_loss: 0.3288 - val_mae_euclidean: 1.1481 - val_jacard_coef: 0.5051
Epoch 159/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2282 - mae_euclidean: 0.7770 - jacard_coef: 0.6286 - val_loss: 0.3229 - val_mae_euclidean: 1.1743 - val_jacard_coef: 0.5118
Epoch 160/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2243 - mae_euclidean: 0.8462 - jacard_coef: 0.6337 - val_loss: 0.3185 - val_mae_euclidean: 1.1317 - val_jacard_coef: 0.5169
Epoch 161/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2202 - mae_euclidean: 0.7995 - jacard_coef: 0.6392 - val_loss: 0.3170 - val_mae_euclidean: 1.1560 - val_jacard_coef: 0.5186
Epoch 162/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2156 - mae_euclidean: 0.6882 - jacard_coef: 0.6454 - val_loss: 0.3211 - val_mae_euclidean: 1.4405 - val_jacard_coef: 0.5139
Epoch 163/200
4/4 [==============================] - 2s 487ms/step - loss: 0.2155 - mae_euclidean: 0.8042 - jacard_coef: 0.6455 - val_loss: 0.3325 - val_mae_euclidean: 1.3346 - val_jacard_coef: 0.5009
Epoch 164/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2155 - mae_euclidean: 0.7525 - jacard_coef: 0.6454 - val_loss: 0.3107 - val_mae_euclidean: 1.2859 - val_jacard_coef: 0.5259
Epoch 165/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2138 - mae_euclidean: 0.7174 - jacard_coef: 0.6478 - val_loss: 0.3116 - val_mae_euclidean: 1.2369 - val_jacard_coef: 0.5249
Epoch 166/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2130 - mae_euclidean: 0.7729 - jacard_coef: 0.6488 - val_loss: 0.3177 - val_mae_euclidean: 1.2143 - val_jacard_coef: 0.5178
Epoch 167/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2154 - mae_euclidean: 0.7856 - jacard_coef: 0.6457 - val_loss: 0.3298 - val_mae_euclidean: 1.6428 - val_jacard_coef: 0.5040
Epoch 168/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2201 - mae_euclidean: 0.9091 - jacard_coef: 0.6394 - val_loss: 0.3280 - val_mae_euclidean: 1.8519 - val_jacard_coef: 0.5060
Epoch 169/200
4/4 [==============================] - 2s 487ms/step - loss: 0.2149 - mae_euclidean: 0.7518 - jacard_coef: 0.6464 - val_loss: 0.3281 - val_mae_euclidean: 1.9307 - val_jacard_coef: 0.5059
Epoch 170/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2144 - mae_euclidean: 0.7051 - jacard_coef: 0.6470 - val_loss: 0.2990 - val_mae_euclidean: 1.6582 - val_jacard_coef: 0.5397
Epoch 171/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2132 - mae_euclidean: 0.8076 - jacard_coef: 0.6487 - val_loss: 0.2982 - val_mae_euclidean: 1.5595 - val_jacard_coef: 0.5406
Epoch 172/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2267 - mae_euclidean: 0.8468 - jacard_coef: 0.6305 - val_loss: 0.3344 - val_mae_euclidean: 1.8893 - val_jacard_coef: 0.4988
Epoch 173/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2108 - mae_euclidean: 0.7051 - jacard_coef: 0.6519 - val_loss: 0.3186 - val_mae_euclidean: 1.5555 - val_jacard_coef: 0.5168
Epoch 174/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2082 - mae_euclidean: 0.6649 - jacard_coef: 0.6555 - val_loss: 0.3150 - val_mae_euclidean: 1.3060 - val_jacard_coef: 0.5209
Epoch 175/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2073 - mae_euclidean: 0.8185 - jacard_coef: 0.6567 - val_loss: 0.3110 - val_mae_euclidean: 1.2852 - val_jacard_coef: 0.5256
Epoch 176/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2115 - mae_euclidean: 0.7972 - jacard_coef: 0.6512 - val_loss: 0.3038 - val_mae_euclidean: 1.1381 - val_jacard_coef: 0.5340
Epoch 177/200
4/4 [==============================] - 2s 493ms/step - loss: 0.2096 - mae_euclidean: 0.7901 - jacard_coef: 0.6535 - val_loss: 0.3349 - val_mae_euclidean: 1.7262 - val_jacard_coef: 0.4983
Epoch 178/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2042 - mae_euclidean: 0.8160 - jacard_coef: 0.6610 - val_loss: 0.3182 - val_mae_euclidean: 1.7814 - val_jacard_coef: 0.5172
Epoch 179/200
4/4 [==============================] - 2s 489ms/step - loss: 0.2059 - mae_euclidean: 0.6667 - jacard_coef: 0.6586 - val_loss: 0.3338 - val_mae_euclidean: 2.4323 - val_jacard_coef: 0.4995
Epoch 180/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2043 - mae_euclidean: 0.7230 - jacard_coef: 0.6609 - val_loss: 0.3112 - val_mae_euclidean: 1.6187 - val_jacard_coef: 0.5253
Epoch 181/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2052 - mae_euclidean: 0.7401 - jacard_coef: 0.6595 - val_loss: 0.3224 - val_mae_euclidean: 2.2578 - val_jacard_coef: 0.5125
Epoch 182/200
4/4 [==============================] - 2s 491ms/step - loss: 0.2080 - mae_euclidean: 0.7641 - jacard_coef: 0.6559 - val_loss: 0.3213 - val_mae_euclidean: 1.2278 - val_jacard_coef: 0.5136
Epoch 183/200
4/4 [==============================] - 2s 491ms/step - loss: 0.2130 - mae_euclidean: 0.7163 - jacard_coef: 0.6490 - val_loss: 0.3161 - val_mae_euclidean: 1.4161 - val_jacard_coef: 0.5197
Epoch 184/200
4/4 [==============================] - 2s 492ms/step - loss: 0.2037 - mae_euclidean: 0.7320 - jacard_coef: 0.6617 - val_loss: 0.3182 - val_mae_euclidean: 1.4097 - val_jacard_coef: 0.5172
Epoch 185/200
4/4 [==============================] - 2s 492ms/step - loss: 0.2008 - mae_euclidean: 0.6395 - jacard_coef: 0.6657 - val_loss: 0.3131 - val_mae_euclidean: 2.2281 - val_jacard_coef: 0.5231
Epoch 186/200
4/4 [==============================] - 2s 491ms/step - loss: 0.2004 - mae_euclidean: 0.6047 - jacard_coef: 0.6662 - val_loss: 0.3112 - val_mae_euclidean: 1.2620 - val_jacard_coef: 0.5254
Epoch 187/200
4/4 [==============================] - 2s 492ms/step - loss: 0.2047 - mae_euclidean: 0.8512 - jacard_coef: 0.6602 - val_loss: 0.3064 - val_mae_euclidean: 1.2930 - val_jacard_coef: 0.5310
Epoch 188/200
4/4 [==============================] - 2s 490ms/step - loss: 0.1989 - mae_euclidean: 0.7427 - jacard_coef: 0.6683 - val_loss: 0.3171 - val_mae_euclidean: 1.3153 - val_jacard_coef: 0.5185
Epoch 189/200
4/4 [==============================] - 2s 490ms/step - loss: 0.1979 - mae_euclidean: 0.7128 - jacard_coef: 0.6696 - val_loss: 0.3051 - val_mae_euclidean: 1.2965 - val_jacard_coef: 0.5324
Epoch 190/200
4/4 [==============================] - 2s 489ms/step - loss: 0.1927 - mae_euclidean: 0.6129 - jacard_coef: 0.6770 - val_loss: 0.3289 - val_mae_euclidean: 1.4021 - val_jacard_coef: 0.5050
Epoch 191/200
4/4 [==============================] - 2s 490ms/step - loss: 0.1899 - mae_euclidean: 0.6343 - jacard_coef: 0.6808 - val_loss: 0.3276 - val_mae_euclidean: 1.4166 - val_jacard_coef: 0.5065
Epoch 192/200
4/4 [==============================] - 2s 490ms/step - loss: 0.1924 - mae_euclidean: 0.6873 - jacard_coef: 0.6773 - val_loss: 0.3116 - val_mae_euclidean: 1.3575 - val_jacard_coef: 0.5249
Epoch 193/200
4/4 [==============================] - 2s 488ms/step - loss: 0.1929 - mae_euclidean: 0.7525 - jacard_coef: 0.6766 - val_loss: 0.3201 - val_mae_euclidean: 1.3801 - val_jacard_coef: 0.5150
Epoch 194/200
4/4 [==============================] - 2s 492ms/step - loss: 0.1964 - mae_euclidean: 0.8713 - jacard_coef: 0.6720 - val_loss: 0.3264 - val_mae_euclidean: 1.3174 - val_jacard_coef: 0.5078
Epoch 195/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2125 - mae_euclidean: 1.0465 - jacard_coef: 0.6499 - val_loss: 0.3508 - val_mae_euclidean: 2.5056 - val_jacard_coef: 0.4806
Epoch 196/200
4/4 [==============================] - 2s 490ms/step - loss: 0.2129 - mae_euclidean: 0.8019 - jacard_coef: 0.6493 - val_loss: 0.3201 - val_mae_euclidean: 1.6354 - val_jacard_coef: 0.5150
Epoch 197/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2100 - mae_euclidean: 0.9021 - jacard_coef: 0.6532 - val_loss: 0.3269 - val_mae_euclidean: 1.2411 - val_jacard_coef: 0.5073
Epoch 198/200
4/4 [==============================] - 2s 488ms/step - loss: 0.2019 - mae_euclidean: 1.1415 - jacard_coef: 0.6641 - val_loss: 0.3451 - val_mae_euclidean: 2.2220 - val_jacard_coef: 0.4869
Epoch 199/200
4/4 [==============================] - 2s 488ms/step - loss: 0.1958 - mae_euclidean: 0.7096 - jacard_coef: 0.6726 - val_loss: 0.3542 - val_mae_euclidean: 1.5990 - val_jacard_coef: 0.4769
Epoch 200/200
4/4 [==============================] - 2s 488ms/step - loss: 0.1968 - mae_euclidean: 0.7984 - jacard_coef: 0.6712 - val_loss: 0.3165 - val_mae_euclidean: 1.1171 - val_jacard_coef: 0.5191
Model trained for 382.39845991134644s

plot_history(history_sa_unet2, "SA Unet")

model_sa_unet2.evaluate(X_val, y_val)

1/1 [==============================] - 0s 352ms/step - loss: 0.3165 - mae_euclidean: 1.1171 - jacard_coef: 0.5191

[0.3165448307991028, 1.1170706748962402, 0.5191344618797302]

Small trend for increasing validation loss - a sign for overfitting. The val_mae_euclidean (the objective of the tune) is smaller (good), but the jacard is significantly smaller (not good) than model_sa_unet1.

plot_train_val(model_sa_unet2, 1, 1)

# continue trainning with smaller learning rate
model_sa_unet2a = tf.keras.models.clone_model(model_sa_unet2)
model_sa_unet2a.set_weights(model_sa_unet2.get_weights())
model_sa_unet2a.compile(optimizer=Adam(learning_rate=1e-3),
                         loss = dice_coef_loss,
                         metrics=[mae_euclidean, jacard_coef]
                         )
history_sa_unet2a = model_sa_unet2a.fit(X_train, y_train,
                                    validation_data = (X_val, y_val),
                                    batch_size = 8,
                                    epochs=100
                                   )
model_sa_unet2a.save("2022-02-17 best SA-UNet2a 200+100epochs.hdf5")

Epoch 1/100
4/4 [==============================] - 6s 584ms/step - loss: 0.1958 - mae_euclidean: 0.7189 - jacard_coef: 0.6725 - val_loss: 0.3379 - val_mae_euclidean: 1.1639 - val_jacard_coef: 0.4949
Epoch 2/100
4/4 [==============================] - 2s 487ms/step - loss: 0.1888 - mae_euclidean: 0.6815 - jacard_coef: 0.6825 - val_loss: 0.3262 - val_mae_euclidean: 1.2222 - val_jacard_coef: 0.5081
Epoch 3/100
4/4 [==============================] - 2s 487ms/step - loss: 0.1839 - mae_euclidean: 0.5711 - jacard_coef: 0.6896 - val_loss: 0.3140 - val_mae_euclidean: 1.2105 - val_jacard_coef: 0.5221
Epoch 4/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1813 - mae_euclidean: 0.7065 - jacard_coef: 0.6931 - val_loss: 0.3113 - val_mae_euclidean: 1.2141 - val_jacard_coef: 0.5252
Epoch 5/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1842 - mae_euclidean: 0.7839 - jacard_coef: 0.6890 - val_loss: 0.3112 - val_mae_euclidean: 1.1063 - val_jacard_coef: 0.5254
Epoch 6/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1810 - mae_euclidean: 0.6225 - jacard_coef: 0.6935 - val_loss: 0.3116 - val_mae_euclidean: 1.1226 - val_jacard_coef: 0.5249
Epoch 7/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1860 - mae_euclidean: 0.7409 - jacard_coef: 0.6865 - val_loss: 0.3110 - val_mae_euclidean: 1.2522 - val_jacard_coef: 0.5255
Epoch 8/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1825 - mae_euclidean: 0.7333 - jacard_coef: 0.6916 - val_loss: 0.3113 - val_mae_euclidean: 1.2125 - val_jacard_coef: 0.5252
Epoch 9/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1807 - mae_euclidean: 0.7355 - jacard_coef: 0.6939 - val_loss: 0.3132 - val_mae_euclidean: 1.2119 - val_jacard_coef: 0.5231
Epoch 10/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1750 - mae_euclidean: 0.5542 - jacard_coef: 0.7021 - val_loss: 0.3115 - val_mae_euclidean: 1.1770 - val_jacard_coef: 0.5250
Epoch 11/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1778 - mae_euclidean: 0.6605 - jacard_coef: 0.6983 - val_loss: 0.3123 - val_mae_euclidean: 1.2942 - val_jacard_coef: 0.5241
Epoch 12/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1736 - mae_euclidean: 0.6087 - jacard_coef: 0.7042 - val_loss: 0.3145 - val_mae_euclidean: 1.2991 - val_jacard_coef: 0.5215
Epoch 13/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1781 - mae_euclidean: 0.6521 - jacard_coef: 0.6977 - val_loss: 0.3222 - val_mae_euclidean: 1.3438 - val_jacard_coef: 0.5127
Epoch 14/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1813 - mae_euclidean: 0.6311 - jacard_coef: 0.6932 - val_loss: 0.3239 - val_mae_euclidean: 1.3467 - val_jacard_coef: 0.5107
Epoch 15/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1719 - mae_euclidean: 0.5984 - jacard_coef: 0.7067 - val_loss: 0.3233 - val_mae_euclidean: 1.2683 - val_jacard_coef: 0.5113
Epoch 16/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1739 - mae_euclidean: 0.6056 - jacard_coef: 0.7037 - val_loss: 0.3272 - val_mae_euclidean: 1.2816 - val_jacard_coef: 0.5070
Epoch 17/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1720 - mae_euclidean: 0.6224 - jacard_coef: 0.7066 - val_loss: 0.3291 - val_mae_euclidean: 1.6318 - val_jacard_coef: 0.5048
Epoch 18/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1750 - mae_euclidean: 0.7930 - jacard_coef: 0.7021 - val_loss: 0.3263 - val_mae_euclidean: 1.3613 - val_jacard_coef: 0.5080
Epoch 19/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1734 - mae_euclidean: 0.6843 - jacard_coef: 0.7045 - val_loss: 0.3221 - val_mae_euclidean: 1.3324 - val_jacard_coef: 0.5127
Epoch 20/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1733 - mae_euclidean: 0.6592 - jacard_coef: 0.7047 - val_loss: 0.3221 - val_mae_euclidean: 1.3410 - val_jacard_coef: 0.5128
Epoch 21/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1723 - mae_euclidean: 0.5644 - jacard_coef: 0.7063 - val_loss: 0.3225 - val_mae_euclidean: 1.3571 - val_jacard_coef: 0.5123
Epoch 22/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1687 - mae_euclidean: 0.5793 - jacard_coef: 0.7114 - val_loss: 0.3229 - val_mae_euclidean: 1.3493 - val_jacard_coef: 0.5119
Epoch 23/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1756 - mae_euclidean: 0.7885 - jacard_coef: 0.7014 - val_loss: 0.3221 - val_mae_euclidean: 1.3621 - val_jacard_coef: 0.5127
Epoch 24/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1690 - mae_euclidean: 0.5017 - jacard_coef: 0.7111 - val_loss: 0.3233 - val_mae_euclidean: 1.3709 - val_jacard_coef: 0.5114
Epoch 25/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1714 - mae_euclidean: 0.6286 - jacard_coef: 0.7074 - val_loss: 0.3268 - val_mae_euclidean: 1.6570 - val_jacard_coef: 0.5073
Epoch 26/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1681 - mae_euclidean: 0.5265 - jacard_coef: 0.7122 - val_loss: 0.3269 - val_mae_euclidean: 1.6446 - val_jacard_coef: 0.5072
Epoch 27/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1742 - mae_euclidean: 0.7484 - jacard_coef: 0.7036 - val_loss: 0.3276 - val_mae_euclidean: 1.6376 - val_jacard_coef: 0.5065
Epoch 28/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1787 - mae_euclidean: 0.7831 - jacard_coef: 0.6970 - val_loss: 0.3271 - val_mae_euclidean: 1.6469 - val_jacard_coef: 0.5071
Epoch 29/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1678 - mae_euclidean: 0.5705 - jacard_coef: 0.7127 - val_loss: 0.3281 - val_mae_euclidean: 1.6605 - val_jacard_coef: 0.5059
Epoch 30/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1681 - mae_euclidean: 0.5789 - jacard_coef: 0.7123 - val_loss: 0.3283 - val_mae_euclidean: 1.6877 - val_jacard_coef: 0.5057
Epoch 31/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1694 - mae_euclidean: 0.6383 - jacard_coef: 0.7103 - val_loss: 0.3325 - val_mae_euclidean: 1.6930 - val_jacard_coef: 0.5010
Epoch 32/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1760 - mae_euclidean: 0.6121 - jacard_coef: 0.7008 - val_loss: 0.3369 - val_mae_euclidean: 1.7180 - val_jacard_coef: 0.4960
Epoch 33/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1686 - mae_euclidean: 0.5580 - jacard_coef: 0.7115 - val_loss: 0.3409 - val_mae_euclidean: 1.7446 - val_jacard_coef: 0.4916
Epoch 34/100
4/4 [==============================] - 2s 487ms/step - loss: 0.1700 - mae_euclidean: 0.6017 - jacard_coef: 0.7095 - val_loss: 0.3396 - val_mae_euclidean: 1.7370 - val_jacard_coef: 0.4930
Epoch 35/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1828 - mae_euclidean: 0.8120 - jacard_coef: 0.6915 - val_loss: 0.3342 - val_mae_euclidean: 1.6876 - val_jacard_coef: 0.4990
Epoch 36/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1655 - mae_euclidean: 0.6011 - jacard_coef: 0.7161 - val_loss: 0.3320 - val_mae_euclidean: 1.6778 - val_jacard_coef: 0.5015
Epoch 37/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1659 - mae_euclidean: 0.5269 - jacard_coef: 0.7155 - val_loss: 0.3304 - val_mae_euclidean: 1.6710 - val_jacard_coef: 0.5033
Epoch 38/100
4/4 [==============================] - 2s 492ms/step - loss: 0.1670 - mae_euclidean: 0.5522 - jacard_coef: 0.7140 - val_loss: 0.3315 - val_mae_euclidean: 1.6708 - val_jacard_coef: 0.5021
Epoch 39/100
4/4 [==============================] - 2s 492ms/step - loss: 0.1652 - mae_euclidean: 0.5408 - jacard_coef: 0.7165 - val_loss: 0.3293 - val_mae_euclidean: 1.6531 - val_jacard_coef: 0.5045
Epoch 40/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1676 - mae_euclidean: 0.6855 - jacard_coef: 0.7130 - val_loss: 0.3303 - val_mae_euclidean: 1.6474 - val_jacard_coef: 0.5034
Epoch 41/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1680 - mae_euclidean: 0.6130 - jacard_coef: 0.7125 - val_loss: 0.3321 - val_mae_euclidean: 1.6357 - val_jacard_coef: 0.5014
Epoch 42/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1683 - mae_euclidean: 0.6119 - jacard_coef: 0.7119 - val_loss: 0.3340 - val_mae_euclidean: 1.6375 - val_jacard_coef: 0.4993
Epoch 43/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1648 - mae_euclidean: 0.5396 - jacard_coef: 0.7171 - val_loss: 0.3371 - val_mae_euclidean: 1.6538 - val_jacard_coef: 0.4958
Epoch 44/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1673 - mae_euclidean: 0.6650 - jacard_coef: 0.7136 - val_loss: 0.3343 - val_mae_euclidean: 1.6512 - val_jacard_coef: 0.4989
Epoch 45/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1656 - mae_euclidean: 0.6421 - jacard_coef: 0.7160 - val_loss: 0.3302 - val_mae_euclidean: 1.6247 - val_jacard_coef: 0.5036
Epoch 46/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1738 - mae_euclidean: 0.6574 - jacard_coef: 0.7041 - val_loss: 0.3302 - val_mae_euclidean: 1.6432 - val_jacard_coef: 0.5035
Epoch 47/100
4/4 [==============================] - 2s 492ms/step - loss: 0.1617 - mae_euclidean: 0.5518 - jacard_coef: 0.7217 - val_loss: 0.3346 - val_mae_euclidean: 1.6579 - val_jacard_coef: 0.4985
Epoch 48/100
4/4 [==============================] - 2s 492ms/step - loss: 0.1634 - mae_euclidean: 0.6316 - jacard_coef: 0.7191 - val_loss: 0.3344 - val_mae_euclidean: 1.7015 - val_jacard_coef: 0.4988
Epoch 49/100
4/4 [==============================] - 2s 492ms/step - loss: 0.1652 - mae_euclidean: 0.5391 - jacard_coef: 0.7165 - val_loss: 0.3394 - val_mae_euclidean: 1.7382 - val_jacard_coef: 0.4932
Epoch 50/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1638 - mae_euclidean: 0.6198 - jacard_coef: 0.7186 - val_loss: 0.3398 - val_mae_euclidean: 1.7354 - val_jacard_coef: 0.4927
Epoch 51/100
4/4 [==============================] - 2s 492ms/step - loss: 0.1671 - mae_euclidean: 0.6439 - jacard_coef: 0.7138 - val_loss: 0.3354 - val_mae_euclidean: 1.6973 - val_jacard_coef: 0.4977
Epoch 52/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1622 - mae_euclidean: 0.5829 - jacard_coef: 0.7210 - val_loss: 0.3277 - val_mae_euclidean: 1.6351 - val_jacard_coef: 0.5063
Epoch 53/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1643 - mae_euclidean: 0.5437 - jacard_coef: 0.7178 - val_loss: 0.3281 - val_mae_euclidean: 1.6529 - val_jacard_coef: 0.5059
Epoch 54/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1624 - mae_euclidean: 0.6079 - jacard_coef: 0.7206 - val_loss: 0.3317 - val_mae_euclidean: 1.6725 - val_jacard_coef: 0.5019
Epoch 55/100
4/4 [==============================] - 2s 492ms/step - loss: 0.1611 - mae_euclidean: 0.5299 - jacard_coef: 0.7225 - val_loss: 0.3311 - val_mae_euclidean: 1.6817 - val_jacard_coef: 0.5026
Epoch 56/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1640 - mae_euclidean: 0.6048 - jacard_coef: 0.7183 - val_loss: 0.3305 - val_mae_euclidean: 1.5832 - val_jacard_coef: 0.5031
Epoch 57/100
4/4 [==============================] - 2s 492ms/step - loss: 0.1614 - mae_euclidean: 0.5357 - jacard_coef: 0.7221 - val_loss: 0.3312 - val_mae_euclidean: 1.5755 - val_jacard_coef: 0.5024
Epoch 58/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1636 - mae_euclidean: 0.5152 - jacard_coef: 0.7188 - val_loss: 0.3307 - val_mae_euclidean: 1.6415 - val_jacard_coef: 0.5030
Epoch 59/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1625 - mae_euclidean: 0.6006 - jacard_coef: 0.7204 - val_loss: 0.3329 - val_mae_euclidean: 1.6043 - val_jacard_coef: 0.5005
Epoch 60/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1701 - mae_euclidean: 0.6786 - jacard_coef: 0.7093 - val_loss: 0.3339 - val_mae_euclidean: 1.6569 - val_jacard_coef: 0.4994
Epoch 61/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1608 - mae_euclidean: 0.5383 - jacard_coef: 0.7230 - val_loss: 0.3326 - val_mae_euclidean: 1.5860 - val_jacard_coef: 0.5009
Epoch 62/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1625 - mae_euclidean: 0.6031 - jacard_coef: 0.7204 - val_loss: 0.3299 - val_mae_euclidean: 1.6330 - val_jacard_coef: 0.5039
Epoch 63/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1702 - mae_euclidean: 0.7025 - jacard_coef: 0.7092 - val_loss: 0.3319 - val_mae_euclidean: 1.6991 - val_jacard_coef: 0.5016
Epoch 64/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1658 - mae_euclidean: 0.6919 - jacard_coef: 0.7157 - val_loss: 0.3374 - val_mae_euclidean: 1.7403 - val_jacard_coef: 0.4954
Epoch 65/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1592 - mae_euclidean: 0.5489 - jacard_coef: 0.7253 - val_loss: 0.3287 - val_mae_euclidean: 1.5947 - val_jacard_coef: 0.5053
Epoch 66/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1712 - mae_euclidean: 0.6109 - jacard_coef: 0.7077 - val_loss: 0.3244 - val_mae_euclidean: 1.5488 - val_jacard_coef: 0.5101
Epoch 67/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1622 - mae_euclidean: 0.5683 - jacard_coef: 0.7210 - val_loss: 0.3290 - val_mae_euclidean: 1.5654 - val_jacard_coef: 0.5049
Epoch 68/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1632 - mae_euclidean: 0.5779 - jacard_coef: 0.7194 - val_loss: 0.3320 - val_mae_euclidean: 1.6453 - val_jacard_coef: 0.5015
Epoch 69/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1605 - mae_euclidean: 0.5989 - jacard_coef: 0.7236 - val_loss: 0.3274 - val_mae_euclidean: 1.6040 - val_jacard_coef: 0.5067
Epoch 70/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1585 - mae_euclidean: 0.5308 - jacard_coef: 0.7264 - val_loss: 0.3276 - val_mae_euclidean: 1.2616 - val_jacard_coef: 0.5065
Epoch 71/100
4/4 [==============================] - 2s 488ms/step - loss: 0.1627 - mae_euclidean: 0.6608 - jacard_coef: 0.7201 - val_loss: 0.3300 - val_mae_euclidean: 1.3000 - val_jacard_coef: 0.5038
Epoch 72/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1629 - mae_euclidean: 0.6057 - jacard_coef: 0.7199 - val_loss: 0.3339 - val_mae_euclidean: 1.3236 - val_jacard_coef: 0.4994
Epoch 73/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1591 - mae_euclidean: 0.4863 - jacard_coef: 0.7254 - val_loss: 0.3352 - val_mae_euclidean: 1.5737 - val_jacard_coef: 0.4979
Epoch 74/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1641 - mae_euclidean: 0.6466 - jacard_coef: 0.7181 - val_loss: 0.3343 - val_mae_euclidean: 1.5730 - val_jacard_coef: 0.4989
Epoch 75/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1578 - mae_euclidean: 0.5578 - jacard_coef: 0.7274 - val_loss: 0.3339 - val_mae_euclidean: 1.5794 - val_jacard_coef: 0.4994
Epoch 76/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1603 - mae_euclidean: 0.5632 - jacard_coef: 0.7238 - val_loss: 0.3339 - val_mae_euclidean: 1.5576 - val_jacard_coef: 0.4994
Epoch 77/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1651 - mae_euclidean: 0.7274 - jacard_coef: 0.7167 - val_loss: 0.3382 - val_mae_euclidean: 1.6457 - val_jacard_coef: 0.4946
Epoch 78/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1633 - mae_euclidean: 0.6357 - jacard_coef: 0.7193 - val_loss: 0.3368 - val_mae_euclidean: 1.6514 - val_jacard_coef: 0.4962
Epoch 79/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1599 - mae_euclidean: 0.5965 - jacard_coef: 0.7243 - val_loss: 0.3355 - val_mae_euclidean: 1.6587 - val_jacard_coef: 0.4976
Epoch 80/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1627 - mae_euclidean: 0.6293 - jacard_coef: 0.7202 - val_loss: 0.3343 - val_mae_euclidean: 1.3853 - val_jacard_coef: 0.4990
Epoch 81/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1627 - mae_euclidean: 0.5265 - jacard_coef: 0.7201 - val_loss: 0.3315 - val_mae_euclidean: 1.3670 - val_jacard_coef: 0.5021
Epoch 82/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1592 - mae_euclidean: 0.6073 - jacard_coef: 0.7254 - val_loss: 0.3314 - val_mae_euclidean: 1.3359 - val_jacard_coef: 0.5022
Epoch 83/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1605 - mae_euclidean: 0.5758 - jacard_coef: 0.7235 - val_loss: 0.3299 - val_mae_euclidean: 1.3340 - val_jacard_coef: 0.5039
Epoch 84/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1598 - mae_euclidean: 0.5041 - jacard_coef: 0.7246 - val_loss: 0.3248 - val_mae_euclidean: 1.3433 - val_jacard_coef: 0.5097
Epoch 85/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1602 - mae_euclidean: 0.5661 - jacard_coef: 0.7238 - val_loss: 0.3299 - val_mae_euclidean: 1.2402 - val_jacard_coef: 0.5038
Epoch 86/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1613 - mae_euclidean: 0.5440 - jacard_coef: 0.7223 - val_loss: 0.3357 - val_mae_euclidean: 1.3729 - val_jacard_coef: 0.4974
Epoch 87/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1588 - mae_euclidean: 0.5525 - jacard_coef: 0.7260 - val_loss: 0.3378 - val_mae_euclidean: 1.6419 - val_jacard_coef: 0.4950
Epoch 88/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1581 - mae_euclidean: 0.5868 - jacard_coef: 0.7271 - val_loss: 0.3375 - val_mae_euclidean: 1.6283 - val_jacard_coef: 0.4953
Epoch 89/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1621 - mae_euclidean: 0.6886 - jacard_coef: 0.7212 - val_loss: 0.3328 - val_mae_euclidean: 1.6288 - val_jacard_coef: 0.5006
Epoch 90/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1575 - mae_euclidean: 0.5595 - jacard_coef: 0.7279 - val_loss: 0.3282 - val_mae_euclidean: 1.6314 - val_jacard_coef: 0.5058
Epoch 91/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1584 - mae_euclidean: 0.5733 - jacard_coef: 0.7266 - val_loss: 0.3271 - val_mae_euclidean: 1.6277 - val_jacard_coef: 0.5070
Epoch 92/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1570 - mae_euclidean: 0.6121 - jacard_coef: 0.7286 - val_loss: 0.3324 - val_mae_euclidean: 1.6347 - val_jacard_coef: 0.5011
Epoch 93/100
4/4 [==============================] - 2s 491ms/step - loss: 0.1619 - mae_euclidean: 0.5987 - jacard_coef: 0.7215 - val_loss: 0.3365 - val_mae_euclidean: 1.6368 - val_jacard_coef: 0.4964
Epoch 94/100
4/4 [==============================] - 2s 489ms/step - loss: 0.1597 - mae_euclidean: 0.6526 - jacard_coef: 0.7247 - val_loss: 0.3316 - val_mae_euclidean: 1.6240 - val_jacard_coef: 0.5019
Epoch 95/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1702 - mae_euclidean: 0.6795 - jacard_coef: 0.7100 - val_loss: 0.3245 - val_mae_euclidean: 1.6218 - val_jacard_coef: 0.5100
Epoch 96/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1579 - mae_euclidean: 0.5053 - jacard_coef: 0.7273 - val_loss: 0.3277 - val_mae_euclidean: 1.6242 - val_jacard_coef: 0.5063
Epoch 97/100
4/4 [==============================] - 2s 490ms/step - loss: 0.1596 - mae_euclidean: 0.5154 - jacard_coef: 0.7247 - val_loss: 0.3286 - val_mae_euclidean: 1.6347 - val_jacard_coef: 0.5054
Epoch 98/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1579 - mae_euclidean: 0.5974 - jacard_coef: 0.7273 - val_loss: 0.3276 - val_mae_euclidean: 1.6448 - val_jacard_coef: 0.5065
Epoch 99/100
4/4 [==============================] - 2s 493ms/step - loss: 0.1572 - mae_euclidean: 0.5393 - jacard_coef: 0.7283 - val_loss: 0.3305 - val_mae_euclidean: 1.6487 - val_jacard_coef: 0.5032
Epoch 100/100
4/4 [==============================] - 2s 494ms/step - loss: 0.1537 - mae_euclidean: 0.5503 - jacard_coef: 0.7336 - val_loss: 0.3305 - val_mae_euclidean: 1.6478 - val_jacard_coef: 0.5032

plot_history(history_sa_unet2a, "SA Unet2a")

No further improvement achieved with the smaller learning rate. No significant over fitting.

model_sa_unet2a.evaluate(X_val, y_val)

1/1 [==============================] - 0s 350ms/step - loss: 0.3305 - mae_euclidean: 1.6478 - jacard_coef: 0.5032

[0.33050423860549927, 1.6478126049041748, 0.5031962394714355]

SA Unet tuning - loss function

Focal loss function is commonly used for segmentation task. Below a hyper parameters search is performed to find the best values for gamma and the weight for the positive class. Comparison with dice loss will be made.

# finding the optimal focal loss parameters
def model_builder_SA_UNet_2(hp):   
    model = SA_UNet(input_shape, block_size=25, keep_prob=0.8, start_neurons=20)

    hp_gamma = hp.Float('gamma', min_value=0, max_value=5, step=1) # gama 0 is crossentropy
    hp_poss_weight = hp.Float('pos_weight', min_value=1, max_value=3, step=0.5)   
    
    model.compile(optimizer = Adam(learning_rate = 1e-2),
                 loss =  BinaryFocalLoss(gamma=hp_gamma, pos_weight=hp_poss_weight),
                 metrics=[mae_euclidean, jacard_coef])
        
    return model

tuner2 = kt.BayesianOptimization(model_builder_SA_UNet_2,
                                     objective = kt.Objective("val_mae_euclidean", direction='min'), 
                                     max_trials=12,
                                     seed = 14,
                                     directory=os.path.normpath('C:/keras_tuner'),
                                     project_name='2022-02-16_SA_UNet_tune2')

INFO:tensorflow:Reloading Oracle from existing project C:\keras_tuner\2022-02-16_SA_UNet_tune2\oracle.json
INFO:tensorflow:Reloading Tuner from C:\keras_tuner\2022-02-16_SA_UNet_tune2\tuner0.json

tf.random.set_seed(14)
tuner2.search(X_train, y_train,
             validation_data = (X_val, y_val),
             batch_size = 8,
             epochs=200
            ) 

Trial 13 Complete [00h 06m 38s]
val_mae_euclidean: 1.0857945680618286

Best val_mae_euclidean So Far: 0.9302276968955994
Total elapsed time: 00h 13m 21s
INFO:tensorflow:Oracle triggered exit

tuner2.results_summary(num_trials=8)

Results summary
Results in C:\keras_tuner\2022-02-16_SA_UNet_tune2
Showing 8 best trials
Objective(name='val_mae_euclidean', direction='min')
Trial summary
Hyperparameters:
gamma: 0.0
pos_weight: 1.0
Score: 0.9302276968955994
Trial summary
Hyperparameters:
gamma: 0.0
pos_weight: 1.0
Score: 0.9340724945068359
Trial summary
Hyperparameters:
gamma: 0.0
pos_weight: 1.0
Score: 0.9897882342338562
Trial summary
Hyperparameters:
gamma: 2.0
pos_weight: 2.0
Score: 0.9955947399139404
Trial summary
Hyperparameters:
gamma: 1.0
pos_weight: 3.0
Score: 1.0241363048553467
Trial summary
Hyperparameters:
gamma: 0.0
pos_weight: 3.0
Score: 1.0857945680618286
Trial summary
Hyperparameters:
gamma: 0.0
pos_weight: 1.0
Score: 1.1025344133377075
Trial summary
Hyperparameters:
gamma: 5.0
pos_weight: 3.0
Score: 1.1217049360275269

The best metric is achieved with gamma=0 and pos_weight=1. This is essentially binary cross entropy. Top 5 models have similar performance.

# SA Unet3
# changing the loss from dice loss to binary cross entropy
tf.random.set_seed(14)
model_sa_unet3 = SA_UNet(input_shape, block_size=25, keep_prob=0.8, start_neurons=20)
model_sa_unet3.compile(optimizer=Adam(learning_rate=1e-2),
                         loss = "BinaryCrossentropy",
                         metrics=[mae_euclidean, jacard_coef]
                         )
start = time.time()
history_sa_unet3 = model_sa_unet3.fit(X_train, y_train,
                                    validation_data = (X_val, y_val),
                                    batch_size = 8, # no resources for 16
                                    epochs=200
                                   )
print(f"Model trained for {time.time() - start}s")
model_sa_unet3.save("2022-02-17 SA-UNet3 200epochs.hdf5")

Epoch 1/200
4/4 [==============================] - 6s 591ms/step - loss: 0.6256 - mae_euclidean: 6.5440 - jacard_coef: 0.0696 - val_loss: 35092.6523 - val_mae_euclidean: 6.1062 - val_jacard_coef: 0.0738
Epoch 2/200
4/4 [==============================] - 2s 496ms/step - loss: 0.4085 - mae_euclidean: 20.9610 - jacard_coef: 0.0680 - val_loss: 162299.3281 - val_mae_euclidean: 5.7550 - val_jacard_coef: 0.0660
Epoch 3/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3270 - mae_euclidean: 1152921435887370240.0000 - jacard_coef: 0.0676 - val_loss: 26849.3926 - val_mae_euclidean: 5.8864 - val_jacard_coef: 0.0704
Epoch 4/200
4/4 [==============================] - 2s 494ms/step - loss: 0.2855 - mae_euclidean: 61.1282 - jacard_coef: 0.0700 - val_loss: 1009.2335 - val_mae_euclidean: 5.8648 - val_jacard_coef: 0.0360
Epoch 5/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2583 - mae_euclidean: 33.7621 - jacard_coef: 0.0815 - val_loss: 236.0236 - val_mae_euclidean: 71.5303 - val_jacard_coef: 0.0019
Epoch 6/200
4/4 [==============================] - 2s 494ms/step - loss: 0.2387 - mae_euclidean: 22.4661 - jacard_coef: 0.0954 - val_loss: 84.0318 - val_mae_euclidean: 87.3181 - val_jacard_coef: 6.4718e-04
Epoch 7/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2291 - mae_euclidean: 19.2555 - jacard_coef: 0.1027 - val_loss: 27.0562 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 0.0023
Epoch 8/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2213 - mae_euclidean: 19.8906 - jacard_coef: 0.1094 - val_loss: 1.1530 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 0.0056
Epoch 9/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2143 - mae_euclidean: 17.8027 - jacard_coef: 0.1174 - val_loss: 0.6390 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 0.0029
Epoch 10/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2036 - mae_euclidean: 14.2079 - jacard_coef: 0.1322 - val_loss: 0.6827 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.9929e-04
Epoch 11/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1926 - mae_euclidean: 10.4539 - jacard_coef: 0.1454 - val_loss: 0.8109 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.6610e-04
Epoch 12/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1842 - mae_euclidean: 6.5181 - jacard_coef: 0.1661 - val_loss: 0.6768 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.1639e-04
Epoch 13/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1783 - mae_euclidean: 5.5300 - jacard_coef: 0.1757 - val_loss: 0.4896 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 0.0025
Epoch 14/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1746 - mae_euclidean: 5.7167 - jacard_coef: 0.1774 - val_loss: 0.5148 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 0.0030
Epoch 15/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1689 - mae_euclidean: 3.8468 - jacard_coef: 0.1980 - val_loss: 0.6478 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.9320e-04
Epoch 16/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1653 - mae_euclidean: 2.9560 - jacard_coef: 0.2070 - val_loss: 0.8067 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.1102e-04
Epoch 17/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1620 - mae_euclidean: 2.5691 - jacard_coef: 0.2181 - val_loss: 0.8735 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.6140e-05
Epoch 18/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1596 - mae_euclidean: 2.5524 - jacard_coef: 0.2226 - val_loss: 0.8733 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.5849e-05
Epoch 19/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1582 - mae_euclidean: 2.2318 - jacard_coef: 0.2303 - val_loss: 0.9001 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0472e-05
Epoch 20/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1560 - mae_euclidean: 1.9863 - jacard_coef: 0.2357 - val_loss: 0.8603 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.7046e-05
Epoch 21/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1528 - mae_euclidean: 2.1065 - jacard_coef: 0.2343 - val_loss: 0.8769 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.3658e-05
Epoch 22/200
4/4 [==============================] - 2s 500ms/step - loss: 0.1513 - mae_euclidean: 1.7357 - jacard_coef: 0.2523 - val_loss: 0.9228 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.7461e-05
Epoch 23/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1534 - mae_euclidean: 1.9145 - jacard_coef: 0.2430 - val_loss: 0.8452 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.1439e-05
Epoch 24/200
4/4 [==============================] - 2s 499ms/step - loss: 0.1502 - mae_euclidean: 1.9503 - jacard_coef: 0.2473 - val_loss: 0.8652 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.6633e-05
Epoch 25/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1494 - mae_euclidean: 1.6708 - jacard_coef: 0.2609 - val_loss: 0.8339 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.4307e-05
Epoch 26/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1479 - mae_euclidean: 1.6500 - jacard_coef: 0.2618 - val_loss: 0.7938 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.9794e-05
Epoch 27/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1474 - mae_euclidean: 1.8681 - jacard_coef: 0.2560 - val_loss: 0.8276 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.6099e-05
Epoch 28/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1488 - mae_euclidean: 1.7329 - jacard_coef: 0.2666 - val_loss: 0.8085 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.3238e-05
Epoch 29/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1454 - mae_euclidean: 1.6367 - jacard_coef: 0.2644 - val_loss: 0.7405 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.1393e-04
Epoch 30/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1455 - mae_euclidean: 1.5351 - jacard_coef: 0.2696 - val_loss: 0.8083 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.4770e-05
Epoch 31/200
4/4 [==============================] - 2s 499ms/step - loss: 0.1441 - mae_euclidean: 1.4306 - jacard_coef: 0.2753 - val_loss: 0.7572 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.0253e-04
Epoch 32/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1450 - mae_euclidean: 1.5643 - jacard_coef: 0.2670 - val_loss: 0.7291 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.2468e-04
Epoch 33/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1421 - mae_euclidean: 1.5171 - jacard_coef: 0.2764 - val_loss: 0.7381 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.1778e-04
Epoch 34/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1413 - mae_euclidean: 1.4653 - jacard_coef: 0.2799 - val_loss: 0.6863 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.8335e-04
Epoch 35/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1450 - mae_euclidean: 1.6530 - jacard_coef: 0.2726 - val_loss: 0.6855 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.8147e-04
Epoch 36/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1395 - mae_euclidean: 1.3924 - jacard_coef: 0.2810 - val_loss: 0.6970 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.6828e-04
Epoch 37/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1382 - mae_euclidean: 1.3452 - jacard_coef: 0.2892 - val_loss: 0.6983 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.6702e-04
Epoch 38/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1386 - mae_euclidean: 1.3330 - jacard_coef: 0.2876 - val_loss: 0.6811 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.9396e-04
Epoch 39/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1376 - mae_euclidean: 1.3754 - jacard_coef: 0.2913 - val_loss: 0.6487 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.6758e-04
Epoch 40/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1382 - mae_euclidean: 1.3759 - jacard_coef: 0.2929 - val_loss: 0.6660 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.3080e-04
Epoch 41/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1365 - mae_euclidean: 1.2951 - jacard_coef: 0.2948 - val_loss: 0.7106 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.6293e-04
Epoch 42/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1378 - mae_euclidean: 1.4395 - jacard_coef: 0.2895 - val_loss: 0.6941 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.9579e-04
Epoch 43/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1351 - mae_euclidean: 1.3715 - jacard_coef: 0.2999 - val_loss: 0.7219 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.5596e-04
Epoch 44/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1357 - mae_euclidean: 1.3668 - jacard_coef: 0.2990 - val_loss: 0.6410 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.2143e-04
Epoch 45/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1351 - mae_euclidean: 1.3911 - jacard_coef: 0.2940 - val_loss: 0.6591 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.7332e-04
Epoch 46/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1365 - mae_euclidean: 1.4248 - jacard_coef: 0.2916 - val_loss: 0.7265 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.6368e-04
Epoch 47/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1337 - mae_euclidean: 1.2800 - jacard_coef: 0.3037 - val_loss: 0.7051 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.9245e-04
Epoch 48/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1335 - mae_euclidean: 1.2185 - jacard_coef: 0.3071 - val_loss: 0.6232 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.0394e-04
Epoch 49/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1339 - mae_euclidean: 1.2195 - jacard_coef: 0.3013 - val_loss: 0.5719 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.3627e-04
Epoch 50/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1322 - mae_euclidean: 1.2003 - jacard_coef: 0.3083 - val_loss: 0.5320 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 0.0012
Epoch 51/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1335 - mae_euclidean: 1.3394 - jacard_coef: 0.3024 - val_loss: 0.5679 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.1619e-04
Epoch 52/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1307 - mae_euclidean: 1.1921 - jacard_coef: 0.3118 - val_loss: 0.5757 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.5939e-04
Epoch 53/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1318 - mae_euclidean: 1.2701 - jacard_coef: 0.3088 - val_loss: 0.5596 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 9.2407e-04
Epoch 54/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1304 - mae_euclidean: 1.3056 - jacard_coef: 0.3153 - val_loss: 0.5023 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 0.0017
Epoch 55/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1298 - mae_euclidean: 1.2364 - jacard_coef: 0.3136 - val_loss: 0.5672 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 9.4360e-04
Epoch 56/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1311 - mae_euclidean: 1.2929 - jacard_coef: 0.3122 - val_loss: 0.5044 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 0.0019
Epoch 57/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1296 - mae_euclidean: 1.2451 - jacard_coef: 0.3197 - val_loss: 0.5082 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 0.0019
Epoch 58/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1291 - mae_euclidean: 1.2361 - jacard_coef: 0.3170 - val_loss: 0.4965 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 0.0025
Epoch 59/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1305 - mae_euclidean: 1.2496 - jacard_coef: 0.3161 - val_loss: 0.4665 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 0.0034
Epoch 60/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1339 - mae_euclidean: 1.2908 - jacard_coef: 0.3113 - val_loss: 0.4385 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 0.0056
Epoch 61/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1302 - mae_euclidean: 1.2032 - jacard_coef: 0.3177 - val_loss: 0.4834 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 0.0046
Epoch 62/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1317 - mae_euclidean: 1.2970 - jacard_coef: 0.3141 - val_loss: 0.4027 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 0.0112
Epoch 63/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1324 - mae_euclidean: 1.3576 - jacard_coef: 0.3093 - val_loss: 0.4244 - val_mae_euclidean: 9223371487098961920.0000 - val_jacard_coef: 0.0085
Epoch 64/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1322 - mae_euclidean: 1.3570 - jacard_coef: 0.3105 - val_loss: 0.4026 - val_mae_euclidean: 6917528477885267968.0000 - val_jacard_coef: 0.0111
Epoch 65/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1290 - mae_euclidean: 1.2493 - jacard_coef: 0.3168 - val_loss: 0.4697 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 0.0057
Epoch 66/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1298 - mae_euclidean: 1.2664 - jacard_coef: 0.3106 - val_loss: 0.4673 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 0.0056
Epoch 67/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1276 - mae_euclidean: 1.1789 - jacard_coef: 0.3238 - val_loss: 0.4268 - val_mae_euclidean: 6917528477885267968.0000 - val_jacard_coef: 0.0095
Epoch 68/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1289 - mae_euclidean: 1.2667 - jacard_coef: 0.3241 - val_loss: 0.4434 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 0.0064
Epoch 69/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1258 - mae_euclidean: 1.2569 - jacard_coef: 0.3307 - val_loss: 0.4766 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0083
Epoch 70/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1249 - mae_euclidean: 1.2185 - jacard_coef: 0.3291 - val_loss: 0.4475 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0086
Epoch 71/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1258 - mae_euclidean: 1.3262 - jacard_coef: 0.3231 - val_loss: 0.4468 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 0.0091
Epoch 72/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1255 - mae_euclidean: 1.2050 - jacard_coef: 0.3346 - val_loss: 0.4421 - val_mae_euclidean: 79.8886 - val_jacard_coef: 0.0121
Epoch 73/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1245 - mae_euclidean: 1.1538 - jacard_coef: 0.3358 - val_loss: 0.4432 - val_mae_euclidean: 73.1379 - val_jacard_coef: 0.0113
Epoch 74/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1261 - mae_euclidean: 1.3606 - jacard_coef: 0.3291 - val_loss: 0.4343 - val_mae_euclidean: 42.5144 - val_jacard_coef: 0.0186
Epoch 75/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1229 - mae_euclidean: 1.2249 - jacard_coef: 0.3370 - val_loss: 0.4883 - val_mae_euclidean: 57.2914 - val_jacard_coef: 0.0113
Epoch 76/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1218 - mae_euclidean: 1.2004 - jacard_coef: 0.3379 - val_loss: 0.4237 - val_mae_euclidean: 34.7500 - val_jacard_coef: 0.0232
Epoch 77/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1225 - mae_euclidean: 1.3101 - jacard_coef: 0.3366 - val_loss: 0.4497 - val_mae_euclidean: 31.9013 - val_jacard_coef: 0.0193
Epoch 78/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1256 - mae_euclidean: 1.2751 - jacard_coef: 0.3374 - val_loss: 0.4445 - val_mae_euclidean: 57.8290 - val_jacard_coef: 0.0128
Epoch 79/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1238 - mae_euclidean: 1.2230 - jacard_coef: 0.3379 - val_loss: 0.3439 - val_mae_euclidean: 16.3804 - val_jacard_coef: 0.0602
Epoch 80/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1241 - mae_euclidean: 1.3246 - jacard_coef: 0.3387 - val_loss: 0.3433 - val_mae_euclidean: 26.0244 - val_jacard_coef: 0.0405
Epoch 81/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1226 - mae_euclidean: 1.2095 - jacard_coef: 0.3365 - val_loss: 0.3949 - val_mae_euclidean: 17.8075 - val_jacard_coef: 0.0406
Epoch 82/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1202 - mae_euclidean: 1.1927 - jacard_coef: 0.3469 - val_loss: 0.3565 - val_mae_euclidean: 27.1684 - val_jacard_coef: 0.0368
Epoch 83/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1212 - mae_euclidean: 1.1469 - jacard_coef: 0.3424 - val_loss: 0.3986 - val_mae_euclidean: 21.6943 - val_jacard_coef: 0.0410
Epoch 84/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1221 - mae_euclidean: 1.1009 - jacard_coef: 0.3485 - val_loss: 0.3873 - val_mae_euclidean: 16.7349 - val_jacard_coef: 0.0420
Epoch 85/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1224 - mae_euclidean: 1.1551 - jacard_coef: 0.3354 - val_loss: 0.3954 - val_mae_euclidean: 20.0871 - val_jacard_coef: 0.0455
Epoch 86/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1192 - mae_euclidean: 1.1334 - jacard_coef: 0.3516 - val_loss: 0.4551 - val_mae_euclidean: 17.0372 - val_jacard_coef: 0.0422
Epoch 87/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1183 - mae_euclidean: 1.1576 - jacard_coef: 0.3546 - val_loss: 0.3272 - val_mae_euclidean: 12.0377 - val_jacard_coef: 0.0772
Epoch 88/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1184 - mae_euclidean: 1.1584 - jacard_coef: 0.3500 - val_loss: 0.4292 - val_mae_euclidean: 17.9212 - val_jacard_coef: 0.0528
Epoch 89/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1204 - mae_euclidean: 1.2424 - jacard_coef: 0.3572 - val_loss: 0.3342 - val_mae_euclidean: 11.3533 - val_jacard_coef: 0.0860
Epoch 90/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1210 - mae_euclidean: 1.2879 - jacard_coef: 0.3446 - val_loss: 0.2918 - val_mae_euclidean: 10.4074 - val_jacard_coef: 0.0982
Epoch 91/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1195 - mae_euclidean: 1.1601 - jacard_coef: 0.3553 - val_loss: 0.4155 - val_mae_euclidean: 11.7489 - val_jacard_coef: 0.0784
Epoch 92/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1175 - mae_euclidean: 1.0857 - jacard_coef: 0.3653 - val_loss: 0.3199 - val_mae_euclidean: 9.5442 - val_jacard_coef: 0.1023
Epoch 93/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1205 - mae_euclidean: 1.1450 - jacard_coef: 0.3416 - val_loss: 0.3770 - val_mae_euclidean: 10.9272 - val_jacard_coef: 0.0943
Epoch 94/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1191 - mae_euclidean: 1.1603 - jacard_coef: 0.3571 - val_loss: 0.3265 - val_mae_euclidean: 8.3780 - val_jacard_coef: 0.1219
Epoch 95/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1240 - mae_euclidean: 1.3945 - jacard_coef: 0.3415 - val_loss: 0.2916 - val_mae_euclidean: 8.8010 - val_jacard_coef: 0.1044
Epoch 96/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1185 - mae_euclidean: 1.1538 - jacard_coef: 0.3476 - val_loss: 0.3570 - val_mae_euclidean: 7.8159 - val_jacard_coef: 0.1229
Epoch 97/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1166 - mae_euclidean: 1.1078 - jacard_coef: 0.3658 - val_loss: 0.3150 - val_mae_euclidean: 7.4873 - val_jacard_coef: 0.1279
Epoch 98/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1174 - mae_euclidean: 1.2434 - jacard_coef: 0.3550 - val_loss: 0.2906 - val_mae_euclidean: 6.7495 - val_jacard_coef: 0.1539
Epoch 99/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1168 - mae_euclidean: 1.0306 - jacard_coef: 0.3614 - val_loss: 0.3051 - val_mae_euclidean: 6.7526 - val_jacard_coef: 0.1444
Epoch 100/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1149 - mae_euclidean: 1.1063 - jacard_coef: 0.3703 - val_loss: 0.2449 - val_mae_euclidean: 5.7634 - val_jacard_coef: 0.1725
Epoch 101/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1173 - mae_euclidean: 1.1071 - jacard_coef: 0.3558 - val_loss: 0.2976 - val_mae_euclidean: 6.7064 - val_jacard_coef: 0.1566
Epoch 102/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1146 - mae_euclidean: 1.0172 - jacard_coef: 0.3712 - val_loss: 0.2904 - val_mae_euclidean: 6.6576 - val_jacard_coef: 0.1585
Epoch 103/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1135 - mae_euclidean: 1.1629 - jacard_coef: 0.3664 - val_loss: 0.2411 - val_mae_euclidean: 5.5556 - val_jacard_coef: 0.2120
Epoch 104/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1144 - mae_euclidean: 1.0767 - jacard_coef: 0.3681 - val_loss: 0.2775 - val_mae_euclidean: 6.4200 - val_jacard_coef: 0.1735
Epoch 105/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1122 - mae_euclidean: 1.0358 - jacard_coef: 0.3745 - val_loss: 0.2394 - val_mae_euclidean: 5.1682 - val_jacard_coef: 0.2261
Epoch 106/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1130 - mae_euclidean: 1.0866 - jacard_coef: 0.3763 - val_loss: 0.2247 - val_mae_euclidean: 5.7362 - val_jacard_coef: 0.2084
Epoch 107/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1131 - mae_euclidean: 1.0414 - jacard_coef: 0.3718 - val_loss: 0.2147 - val_mae_euclidean: 4.9684 - val_jacard_coef: 0.2399
Epoch 108/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1139 - mae_euclidean: 1.0427 - jacard_coef: 0.3731 - val_loss: 0.2487 - val_mae_euclidean: 5.5923 - val_jacard_coef: 0.2237
Epoch 109/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1150 - mae_euclidean: 1.0623 - jacard_coef: 0.3697 - val_loss: 0.1770 - val_mae_euclidean: 3.8719 - val_jacard_coef: 0.2750
Epoch 110/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1106 - mae_euclidean: 1.0860 - jacard_coef: 0.3698 - val_loss: 0.1998 - val_mae_euclidean: 4.1074 - val_jacard_coef: 0.2627
Epoch 111/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1091 - mae_euclidean: 0.9676 - jacard_coef: 0.3857 - val_loss: 0.2509 - val_mae_euclidean: 5.1107 - val_jacard_coef: 0.2366
Epoch 112/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1107 - mae_euclidean: 1.1334 - jacard_coef: 0.3821 - val_loss: 0.2157 - val_mae_euclidean: 4.3528 - val_jacard_coef: 0.2523
Epoch 113/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1090 - mae_euclidean: 1.0095 - jacard_coef: 0.3810 - val_loss: 0.2191 - val_mae_euclidean: 4.6859 - val_jacard_coef: 0.2465
Epoch 114/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1097 - mae_euclidean: 1.0491 - jacard_coef: 0.3836 - val_loss: 0.2197 - val_mae_euclidean: 4.3662 - val_jacard_coef: 0.2671
Epoch 115/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1079 - mae_euclidean: 0.9549 - jacard_coef: 0.3934 - val_loss: 0.2259 - val_mae_euclidean: 4.7127 - val_jacard_coef: 0.2434
Epoch 116/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1066 - mae_euclidean: 0.9700 - jacard_coef: 0.3930 - val_loss: 0.2095 - val_mae_euclidean: 3.8354 - val_jacard_coef: 0.2886
Epoch 117/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1078 - mae_euclidean: 0.9961 - jacard_coef: 0.3929 - val_loss: 0.2054 - val_mae_euclidean: 3.8933 - val_jacard_coef: 0.2794
Epoch 118/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1041 - mae_euclidean: 0.9643 - jacard_coef: 0.4019 - val_loss: 0.2346 - val_mae_euclidean: 4.5963 - val_jacard_coef: 0.2681
Epoch 119/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1062 - mae_euclidean: 0.9205 - jacard_coef: 0.4031 - val_loss: 0.2357 - val_mae_euclidean: 4.9118 - val_jacard_coef: 0.2544
Epoch 120/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1040 - mae_euclidean: 0.9568 - jacard_coef: 0.4056 - val_loss: 0.1926 - val_mae_euclidean: 3.6176 - val_jacard_coef: 0.3059
Epoch 121/200
4/4 [==============================] - 2s 496ms/step - loss: 0.1044 - mae_euclidean: 0.9520 - jacard_coef: 0.4051 - val_loss: 0.2040 - val_mae_euclidean: 3.9242 - val_jacard_coef: 0.2901
Epoch 122/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1033 - mae_euclidean: 0.8500 - jacard_coef: 0.4051 - val_loss: 0.2482 - val_mae_euclidean: 4.8183 - val_jacard_coef: 0.2547
Epoch 123/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1066 - mae_euclidean: 1.0709 - jacard_coef: 0.4021 - val_loss: 0.1979 - val_mae_euclidean: 3.8512 - val_jacard_coef: 0.2852
Epoch 124/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1041 - mae_euclidean: 0.9735 - jacard_coef: 0.4024 - val_loss: 0.1981 - val_mae_euclidean: 3.5437 - val_jacard_coef: 0.2956
Epoch 125/200
4/4 [==============================] - 2s 495ms/step - loss: 0.1023 - mae_euclidean: 0.9186 - jacard_coef: 0.4121 - val_loss: 0.1776 - val_mae_euclidean: 3.1697 - val_jacard_coef: 0.3157
Epoch 126/200
4/4 [==============================] - 2s 494ms/step - loss: 0.1057 - mae_euclidean: 1.1270 - jacard_coef: 0.3933 - val_loss: 0.2116 - val_mae_euclidean: 4.0107 - val_jacard_coef: 0.2855
Epoch 127/200
4/4 [==============================] - 2s 493ms/step - loss: 0.1047 - mae_euclidean: 0.8701 - jacard_coef: 0.4092 - val_loss: 0.2024 - val_mae_euclidean: 3.5487 - val_jacard_coef: 0.3003
Epoch 128/200
4/4 [==============================] - 2s 492ms/step - loss: 0.1033 - mae_euclidean: 1.0535 - jacard_coef: 0.4087 - val_loss: 0.1599 - val_mae_euclidean: 2.8998 - val_jacard_coef: 0.3574
Epoch 129/200
4/4 [==============================] - 2s 493ms/step - loss: 0.1038 - mae_euclidean: 0.9602 - jacard_coef: 0.4115 - val_loss: 0.2356 - val_mae_euclidean: 4.2757 - val_jacard_coef: 0.2652
Epoch 130/200
4/4 [==============================] - 2s 491ms/step - loss: 0.1011 - mae_euclidean: 0.8665 - jacard_coef: 0.4160 - val_loss: 0.1744 - val_mae_euclidean: 3.2847 - val_jacard_coef: 0.3337
Epoch 131/200
4/4 [==============================] - 2s 493ms/step - loss: 0.1012 - mae_euclidean: 0.8721 - jacard_coef: 0.4160 - val_loss: 0.2209 - val_mae_euclidean: 3.9448 - val_jacard_coef: 0.2711
Epoch 132/200
4/4 [==============================] - 2s 493ms/step - loss: 0.1036 - mae_euclidean: 0.9349 - jacard_coef: 0.4116 - val_loss: 0.1816 - val_mae_euclidean: 2.9636 - val_jacard_coef: 0.3319
Epoch 133/200
4/4 [==============================] - 2s 491ms/step - loss: 0.0998 - mae_euclidean: 0.8835 - jacard_coef: 0.4164 - val_loss: 0.1947 - val_mae_euclidean: 3.2206 - val_jacard_coef: 0.3220
Epoch 134/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0964 - mae_euclidean: 0.8091 - jacard_coef: 0.4271 - val_loss: 0.1882 - val_mae_euclidean: 3.1202 - val_jacard_coef: 0.3335
Epoch 135/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0959 - mae_euclidean: 0.7427 - jacard_coef: 0.4371 - val_loss: 0.1870 - val_mae_euclidean: 3.1744 - val_jacard_coef: 0.3328
Epoch 136/200
4/4 [==============================] - 2s 491ms/step - loss: 0.0955 - mae_euclidean: 0.8808 - jacard_coef: 0.4380 - val_loss: 0.1774 - val_mae_euclidean: 3.0453 - val_jacard_coef: 0.3511
Epoch 137/200
4/4 [==============================] - 2s 496ms/step - loss: 0.0942 - mae_euclidean: 0.7919 - jacard_coef: 0.4438 - val_loss: 0.2020 - val_mae_euclidean: 3.2806 - val_jacard_coef: 0.3282
Epoch 138/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0983 - mae_euclidean: 1.1193 - jacard_coef: 0.4301 - val_loss: 0.1728 - val_mae_euclidean: 2.9823 - val_jacard_coef: 0.3525
Epoch 139/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0933 - mae_euclidean: 0.7633 - jacard_coef: 0.4449 - val_loss: 0.2028 - val_mae_euclidean: 3.3662 - val_jacard_coef: 0.3205
Epoch 140/200
4/4 [==============================] - 2s 495ms/step - loss: 0.0933 - mae_euclidean: 0.7365 - jacard_coef: 0.4485 - val_loss: 0.1571 - val_mae_euclidean: 2.7355 - val_jacard_coef: 0.3845
Epoch 141/200
4/4 [==============================] - 2s 495ms/step - loss: 0.0967 - mae_euclidean: 1.2026 - jacard_coef: 0.4344 - val_loss: 0.1736 - val_mae_euclidean: 2.8638 - val_jacard_coef: 0.3508
Epoch 142/200
4/4 [==============================] - 2s 495ms/step - loss: 0.0968 - mae_euclidean: 0.8426 - jacard_coef: 0.4381 - val_loss: 0.1844 - val_mae_euclidean: 3.0428 - val_jacard_coef: 0.3484
Epoch 143/200
4/4 [==============================] - 2s 496ms/step - loss: 0.0931 - mae_euclidean: 0.8085 - jacard_coef: 0.4486 - val_loss: 0.1586 - val_mae_euclidean: 2.2117 - val_jacard_coef: 0.3904
Epoch 144/200
4/4 [==============================] - 2s 496ms/step - loss: 0.0935 - mae_euclidean: 0.8520 - jacard_coef: 0.4499 - val_loss: 0.1906 - val_mae_euclidean: 3.0314 - val_jacard_coef: 0.3398
Epoch 145/200
4/4 [==============================] - 2s 497ms/step - loss: 0.0917 - mae_euclidean: 0.7687 - jacard_coef: 0.4529 - val_loss: 0.1659 - val_mae_euclidean: 2.6672 - val_jacard_coef: 0.3904
Epoch 146/200
4/4 [==============================] - 2s 496ms/step - loss: 0.0906 - mae_euclidean: 0.7637 - jacard_coef: 0.4575 - val_loss: 0.1639 - val_mae_euclidean: 2.7941 - val_jacard_coef: 0.3950
Epoch 147/200
4/4 [==============================] - 2s 496ms/step - loss: 0.0903 - mae_euclidean: 0.7626 - jacard_coef: 0.4649 - val_loss: 0.1463 - val_mae_euclidean: 2.1850 - val_jacard_coef: 0.4220
Epoch 148/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0913 - mae_euclidean: 1.0175 - jacard_coef: 0.4590 - val_loss: 0.1550 - val_mae_euclidean: 2.1106 - val_jacard_coef: 0.3945
Epoch 149/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0948 - mae_euclidean: 0.8760 - jacard_coef: 0.4512 - val_loss: 0.1562 - val_mae_euclidean: 2.2698 - val_jacard_coef: 0.3905
Epoch 150/200
4/4 [==============================] - 2s 495ms/step - loss: 0.0912 - mae_euclidean: 0.8699 - jacard_coef: 0.4553 - val_loss: 0.1262 - val_mae_euclidean: 1.2687 - val_jacard_coef: 0.4507
Epoch 151/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0888 - mae_euclidean: 0.7689 - jacard_coef: 0.4700 - val_loss: 0.1559 - val_mae_euclidean: 2.1265 - val_jacard_coef: 0.3990
Epoch 152/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0886 - mae_euclidean: 0.8342 - jacard_coef: 0.4681 - val_loss: 0.1721 - val_mae_euclidean: 2.8104 - val_jacard_coef: 0.3899
Epoch 153/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0862 - mae_euclidean: 0.6932 - jacard_coef: 0.4774 - val_loss: 0.1278 - val_mae_euclidean: 1.3090 - val_jacard_coef: 0.4550
Epoch 154/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0864 - mae_euclidean: 0.7221 - jacard_coef: 0.4777 - val_loss: 0.1533 - val_mae_euclidean: 2.1849 - val_jacard_coef: 0.4097
Epoch 155/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0856 - mae_euclidean: 0.7242 - jacard_coef: 0.4838 - val_loss: 0.1646 - val_mae_euclidean: 2.6026 - val_jacard_coef: 0.4246
Epoch 156/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0861 - mae_euclidean: 0.6737 - jacard_coef: 0.4809 - val_loss: 0.1857 - val_mae_euclidean: 2.9316 - val_jacard_coef: 0.3799
Epoch 157/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0852 - mae_euclidean: 0.7138 - jacard_coef: 0.4788 - val_loss: 0.1426 - val_mae_euclidean: 1.4227 - val_jacard_coef: 0.4473
Epoch 158/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0893 - mae_euclidean: 0.8421 - jacard_coef: 0.4765 - val_loss: 0.1602 - val_mae_euclidean: 1.9300 - val_jacard_coef: 0.4036
Epoch 159/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0864 - mae_euclidean: 0.7349 - jacard_coef: 0.4807 - val_loss: 0.1518 - val_mae_euclidean: 2.0217 - val_jacard_coef: 0.4237
Epoch 160/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0815 - mae_euclidean: 0.5918 - jacard_coef: 0.4933 - val_loss: 0.1675 - val_mae_euclidean: 2.6831 - val_jacard_coef: 0.4203
Epoch 161/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0813 - mae_euclidean: 0.6209 - jacard_coef: 0.4975 - val_loss: 0.1567 - val_mae_euclidean: 1.8549 - val_jacard_coef: 0.4335
Epoch 162/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0793 - mae_euclidean: 0.7436 - jacard_coef: 0.5082 - val_loss: 0.1497 - val_mae_euclidean: 1.4273 - val_jacard_coef: 0.4494
Epoch 163/200
4/4 [==============================] - 2s 491ms/step - loss: 0.0801 - mae_euclidean: 0.7219 - jacard_coef: 0.5100 - val_loss: 0.1903 - val_mae_euclidean: 2.8102 - val_jacard_coef: 0.3901
Epoch 164/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0781 - mae_euclidean: 0.5588 - jacard_coef: 0.5153 - val_loss: 0.1428 - val_mae_euclidean: 1.2351 - val_jacard_coef: 0.4615
Epoch 165/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0795 - mae_euclidean: 0.6798 - jacard_coef: 0.5160 - val_loss: 0.1692 - val_mae_euclidean: 1.9808 - val_jacard_coef: 0.4206
Epoch 166/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0768 - mae_euclidean: 0.6211 - jacard_coef: 0.5203 - val_loss: 0.1387 - val_mae_euclidean: 1.3454 - val_jacard_coef: 0.4598
Epoch 167/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0774 - mae_euclidean: 0.7417 - jacard_coef: 0.5209 - val_loss: 0.1715 - val_mae_euclidean: 1.9014 - val_jacard_coef: 0.4210
Epoch 168/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0771 - mae_euclidean: 0.6565 - jacard_coef: 0.5221 - val_loss: 0.1470 - val_mae_euclidean: 1.3695 - val_jacard_coef: 0.4574
Epoch 169/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0775 - mae_euclidean: 0.6851 - jacard_coef: 0.5225 - val_loss: 0.1789 - val_mae_euclidean: 2.2386 - val_jacard_coef: 0.4149
Epoch 170/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0762 - mae_euclidean: 0.7423 - jacard_coef: 0.5266 - val_loss: 0.1476 - val_mae_euclidean: 1.2864 - val_jacard_coef: 0.4600
Epoch 171/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0737 - mae_euclidean: 0.6655 - jacard_coef: 0.5360 - val_loss: 0.1650 - val_mae_euclidean: 1.8278 - val_jacard_coef: 0.4401
Epoch 172/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0787 - mae_euclidean: 0.8843 - jacard_coef: 0.5237 - val_loss: 0.1489 - val_mae_euclidean: 1.3762 - val_jacard_coef: 0.4641
Epoch 173/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0726 - mae_euclidean: 0.8047 - jacard_coef: 0.5434 - val_loss: 0.1349 - val_mae_euclidean: 1.0559 - val_jacard_coef: 0.4794
Epoch 174/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0732 - mae_euclidean: 0.5377 - jacard_coef: 0.5449 - val_loss: 0.1514 - val_mae_euclidean: 1.6726 - val_jacard_coef: 0.4485
Epoch 175/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0722 - mae_euclidean: 0.5893 - jacard_coef: 0.5450 - val_loss: 0.1476 - val_mae_euclidean: 1.3248 - val_jacard_coef: 0.4572
Epoch 176/200
4/4 [==============================] - 2s 495ms/step - loss: 0.0729 - mae_euclidean: 0.7713 - jacard_coef: 0.5423 - val_loss: 0.1691 - val_mae_euclidean: 1.4307 - val_jacard_coef: 0.4321
Epoch 177/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0717 - mae_euclidean: 0.6697 - jacard_coef: 0.5471 - val_loss: 0.1536 - val_mae_euclidean: 1.3854 - val_jacard_coef: 0.4609
Epoch 178/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0692 - mae_euclidean: 0.4895 - jacard_coef: 0.5526 - val_loss: 0.1490 - val_mae_euclidean: 1.3071 - val_jacard_coef: 0.4672
Epoch 179/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0698 - mae_euclidean: 0.5738 - jacard_coef: 0.5605 - val_loss: 0.1597 - val_mae_euclidean: 1.3696 - val_jacard_coef: 0.4620
Epoch 180/200
4/4 [==============================] - 2s 495ms/step - loss: 0.0670 - mae_euclidean: 0.5513 - jacard_coef: 0.5702 - val_loss: 0.1500 - val_mae_euclidean: 1.3538 - val_jacard_coef: 0.4783
Epoch 181/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0669 - mae_euclidean: 0.4366 - jacard_coef: 0.5712 - val_loss: 0.1704 - val_mae_euclidean: 1.3771 - val_jacard_coef: 0.4434
Epoch 182/200
4/4 [==============================] - 2s 495ms/step - loss: 0.0668 - mae_euclidean: 0.5525 - jacard_coef: 0.5719 - val_loss: 0.1666 - val_mae_euclidean: 1.3508 - val_jacard_coef: 0.4642
Epoch 183/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0654 - mae_euclidean: 0.5104 - jacard_coef: 0.5774 - val_loss: 0.1620 - val_mae_euclidean: 1.4071 - val_jacard_coef: 0.4648
Epoch 184/200
4/4 [==============================] - 2s 491ms/step - loss: 0.0638 - mae_euclidean: 0.5257 - jacard_coef: 0.5837 - val_loss: 0.1693 - val_mae_euclidean: 1.7664 - val_jacard_coef: 0.4555
Epoch 185/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0632 - mae_euclidean: 0.4176 - jacard_coef: 0.5900 - val_loss: 0.1572 - val_mae_euclidean: 1.4163 - val_jacard_coef: 0.4690
Epoch 186/200
4/4 [==============================] - 2s 495ms/step - loss: 0.0622 - mae_euclidean: 0.3982 - jacard_coef: 0.5937 - val_loss: 0.1486 - val_mae_euclidean: 1.2644 - val_jacard_coef: 0.4771
Epoch 187/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0650 - mae_euclidean: 0.5350 - jacard_coef: 0.5852 - val_loss: 0.1592 - val_mae_euclidean: 1.2668 - val_jacard_coef: 0.4654
Epoch 188/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0639 - mae_euclidean: 0.4657 - jacard_coef: 0.5911 - val_loss: 0.1580 - val_mae_euclidean: 1.4522 - val_jacard_coef: 0.4576
Epoch 189/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0627 - mae_euclidean: 0.4975 - jacard_coef: 0.5915 - val_loss: 0.1612 - val_mae_euclidean: 1.3866 - val_jacard_coef: 0.4664
Epoch 190/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0602 - mae_euclidean: 0.3724 - jacard_coef: 0.6046 - val_loss: 0.1681 - val_mae_euclidean: 1.3677 - val_jacard_coef: 0.4601
Epoch 191/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0603 - mae_euclidean: 0.5073 - jacard_coef: 0.6044 - val_loss: 0.1590 - val_mae_euclidean: 1.3443 - val_jacard_coef: 0.4758
Epoch 192/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0605 - mae_euclidean: 0.4058 - jacard_coef: 0.6072 - val_loss: 0.1684 - val_mae_euclidean: 1.3877 - val_jacard_coef: 0.4607
Epoch 193/200
4/4 [==============================] - 2s 493ms/step - loss: 0.0601 - mae_euclidean: 0.4223 - jacard_coef: 0.6087 - val_loss: 0.1710 - val_mae_euclidean: 1.6969 - val_jacard_coef: 0.4711
Epoch 194/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0603 - mae_euclidean: 0.5903 - jacard_coef: 0.6048 - val_loss: 0.1562 - val_mae_euclidean: 1.2121 - val_jacard_coef: 0.4814
Epoch 195/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0649 - mae_euclidean: 0.7150 - jacard_coef: 0.5894 - val_loss: 0.1764 - val_mae_euclidean: 1.7122 - val_jacard_coef: 0.4658
Epoch 196/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0649 - mae_euclidean: 0.6689 - jacard_coef: 0.5954 - val_loss: 0.1592 - val_mae_euclidean: 1.1036 - val_jacard_coef: 0.4696
Epoch 197/200
4/4 [==============================] - 2s 492ms/step - loss: 0.0674 - mae_euclidean: 0.6684 - jacard_coef: 0.5772 - val_loss: 0.1819 - val_mae_euclidean: 1.5406 - val_jacard_coef: 0.4522
Epoch 198/200
4/4 [==============================] - 2s 491ms/step - loss: 0.0656 - mae_euclidean: 0.7373 - jacard_coef: 0.5861 - val_loss: 0.1544 - val_mae_euclidean: 1.2252 - val_jacard_coef: 0.4813
Epoch 199/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0620 - mae_euclidean: 0.4680 - jacard_coef: 0.5973 - val_loss: 0.1540 - val_mae_euclidean: 1.2381 - val_jacard_coef: 0.4800
Epoch 200/200
4/4 [==============================] - 2s 494ms/step - loss: 0.0615 - mae_euclidean: 0.6253 - jacard_coef: 0.5991 - val_loss: 0.1613 - val_mae_euclidean: 1.2337 - val_jacard_coef: 0.4869
Model trained for 395.1184892654419s

plot_history(history_sa_unet3, "SA Unet3")

plot_train_val(model_sa_unet3, 1, 1)

The binary cross entropy produced similar MAE euclidean and a smaller jacard loss. Loss is approaching 0, limiting the space for improvement. The risk for overfitting with binary cross entropy is higher.

SA Unet tuning - Adam parameters

# finding the optimal Adam seetings
def model_builder_SA_UNet_3(hp):   
    model = SA_UNet(input_shape, block_size=25, keep_prob=0.8, start_neurons=20)

    hp_learning_rate = hp.Float('learning_rate', min_value=0.001, max_value=0.01, step=(0.01-0.001))  
    hp_beta_1 = hp.Float('beta_1', min_value=0.85, max_value=0.95, step=0.01)  
    hp_beta_2 = hp.Float('beta_2', min_value=0.9, max_value=0.999, step=0.001)   
    
    model.compile(optimizer = Adam(learning_rate = hp_learning_rate, beta_1 = hp_beta_1, beta_2 = hp_beta_2),
                 loss = dice_coef_loss,
                 metrics=[mae_euclidean, jacard_coef])
        
    return model

tuner3 = kt.BayesianOptimization(model_builder_SA_UNet_3,
                                     objective = kt.Objective("val_mae_euclidean", direction='min'), 
                                     max_trials=10,
                                     seed = 14,
                                     directory=os.path.normpath('C:/keras_tuner'),
                                     project_name='2022-02-16_SA_UNet_tune3')

INFO:tensorflow:Reloading Oracle from existing project C:\keras_tuner\2022-02-16_SA_UNet_tune3\oracle.json
INFO:tensorflow:Reloading Tuner from C:\keras_tuner\2022-02-16_SA_UNet_tune3\tuner0.json

tf.random.set_seed(14)
tuner3.search(X_train, y_train,
             validation_data = (X_val, y_val),
             batch_size = 8,
             epochs=200
            ) 

INFO:tensorflow:Oracle triggered exit

tuner3.results_summary(num_trials=10)

Results summary
Results in C:\keras_tuner\2022-02-16_SA_UNet_tune3
Showing 10 best trials
Objective(name='val_mae_euclidean', direction='min')
Trial summary
Hyperparameters:
learning_rate: 0.010000000000000002
beta_1: 0.85
beta_2: 0.919
Score: 0.8949853777885437
Trial summary
Hyperparameters:
learning_rate: 0.010000000000000002
beta_1: 0.85
beta_2: 0.9
Score: 0.8990997076034546
Trial summary
Hyperparameters:
learning_rate: 0.010000000000000002
beta_1: 0.85
beta_2: 0.9530000000000001
Score: 0.9050353765487671
Trial summary
Hyperparameters:
learning_rate: 0.010000000000000002
beta_1: 0.85
beta_2: 0.914
Score: 0.9506241083145142
Trial summary
Hyperparameters:
learning_rate: 0.010000000000000002
beta_1: 0.85
beta_2: 0.937
Score: 0.9784936308860779
Trial summary
Hyperparameters:
learning_rate: 0.010000000000000002
beta_1: 0.87
beta_2: 0.9670000000000001
Score: 1.0058289766311646
Trial summary
Hyperparameters:
learning_rate: 0.010000000000000002
beta_1: 0.85
beta_2: 0.9990000000000001
Score: 1.0201386213302612
Trial summary
Hyperparameters:
learning_rate: 0.010000000000000002
beta_1: 0.92
beta_2: 0.9
Score: 1.035428762435913
Trial summary
Hyperparameters:
learning_rate: 0.001
beta_1: 0.85
beta_2: 0.9
Score: 1.091120958328247
Trial summary
Hyperparameters:
learning_rate: 0.001
beta_1: 0.9500000000000001
beta_2: 0.936
Score: 1.131001353263855

The best parameters differ from the already used, but the benefits are small/negligible. All top settings use beta_1=0.85, which is on the edge of the search space; lower beta_1 might be beneficial, but not be explored in this work. Let's train and look at the curves.

# SA Unet4
# changing the Adam parameters learning_rate=1e-2, beta_1=0.85, beta_2=0.919
tf.random.set_seed(14)
model_sa_unet4 = SA_UNet(input_shape, block_size=25, keep_prob=0.8, start_neurons=20)
model_sa_unet4.compile(optimizer=Adam(learning_rate=1e-2, beta_1=0.85, beta_2=0.919),
                         loss = dice_coef_loss,
                         metrics=[mae_euclidean, jacard_coef]
                         )
start = time.time()
history_sa_unet4 = model_sa_unet4.fit(X_train, y_train,
                                    validation_data = (X_val, y_val),
                                    batch_size = 8, # no resources for 16
                                    epochs=200
                                   )
print(f"Model trained for {time.time() - start}s")
model_sa_unet4.save("2022-02-17 SA-UNet4 200epochs.hdf5")

Epoch 1/200
4/4 [==============================] - 5s 583ms/step - loss: 0.8377 - mae_euclidean: 5.3094 - jacard_coef: 0.0892 - val_loss: 0.8610 - val_mae_euclidean: 5.9715 - val_jacard_coef: 0.0747
Epoch 2/200
4/4 [==============================] - 2s 485ms/step - loss: 0.7669 - mae_euclidean: 3.6656 - jacard_coef: 0.1322 - val_loss: 0.9999 - val_mae_euclidean: 159.4866 - val_jacard_coef: 6.1341e-05
Epoch 3/200
4/4 [==============================] - 2s 485ms/step - loss: 0.7218 - mae_euclidean: 3.2920 - jacard_coef: 0.1616 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.4674e-05
Epoch 4/200
4/4 [==============================] - 2s 485ms/step - loss: 0.6825 - mae_euclidean: 3.0739 - jacard_coef: 0.1888 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.6861e-05
Epoch 5/200
4/4 [==============================] - 2s 487ms/step - loss: 0.6443 - mae_euclidean: 2.7700 - jacard_coef: 0.2171 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.6861e-05
Epoch 6/200
4/4 [==============================] - 2s 487ms/step - loss: 0.5993 - mae_euclidean: 2.4325 - jacard_coef: 0.2507 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.7006e-05
Epoch 7/200
4/4 [==============================] - 2s 489ms/step - loss: 0.5659 - mae_euclidean: 2.0500 - jacard_coef: 0.2781 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.6868e-05
Epoch 8/200
4/4 [==============================] - 2s 488ms/step - loss: 0.5248 - mae_euclidean: 1.9873 - jacard_coef: 0.3121 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.8294e-05
Epoch 9/200
4/4 [==============================] - 2s 486ms/step - loss: 0.4926 - mae_euclidean: 1.7374 - jacard_coef: 0.3403 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.2723e-05
Epoch 10/200
4/4 [==============================] - 2s 488ms/step - loss: 0.4605 - mae_euclidean: 1.4598 - jacard_coef: 0.3694 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.1314e-05
Epoch 11/200
4/4 [==============================] - 2s 488ms/step - loss: 0.4427 - mae_euclidean: 1.4173 - jacard_coef: 0.3866 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.8128e-05
Epoch 12/200
4/4 [==============================] - 2s 490ms/step - loss: 0.4223 - mae_euclidean: 1.4714 - jacard_coef: 0.4062 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.0797e-05
Epoch 13/200
4/4 [==============================] - 2s 492ms/step - loss: 0.4179 - mae_euclidean: 1.5727 - jacard_coef: 0.4107 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.0164e-05
Epoch 14/200
4/4 [==============================] - 2s 492ms/step - loss: 0.4186 - mae_euclidean: 1.3531 - jacard_coef: 0.4100 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.7651e-05
Epoch 15/200
4/4 [==============================] - 2s 491ms/step - loss: 0.4008 - mae_euclidean: 1.6014 - jacard_coef: 0.4278 - val_loss: 0.9992 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.9767e-04
Epoch 16/200
4/4 [==============================] - 2s 491ms/step - loss: 0.3917 - mae_euclidean: 1.4183 - jacard_coef: 0.4371 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.0905e-05
Epoch 17/200
4/4 [==============================] - 2s 491ms/step - loss: 0.3776 - mae_euclidean: 1.3859 - jacard_coef: 0.4519 - val_loss: 0.9995 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.9004e-04
Epoch 18/200
4/4 [==============================] - 2s 493ms/step - loss: 0.3870 - mae_euclidean: 1.3800 - jacard_coef: 0.4420 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.0939e-05
Epoch 19/200
4/4 [==============================] - 2s 491ms/step - loss: 0.3734 - mae_euclidean: 1.6324 - jacard_coef: 0.4564 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.8701e-05
Epoch 20/200
4/4 [==============================] - 2s 493ms/step - loss: 0.3715 - mae_euclidean: 1.2398 - jacard_coef: 0.4584 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.1128e-04
Epoch 21/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3626 - mae_euclidean: 1.4462 - jacard_coef: 0.4681 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 9.7776e-05
Epoch 22/200
4/4 [==============================] - 2s 496ms/step - loss: 0.3514 - mae_euclidean: 1.2008 - jacard_coef: 0.4800 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.0951e-04
Epoch 23/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3707 - mae_euclidean: 1.2854 - jacard_coef: 0.4600 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0930e-05
Epoch 24/200
4/4 [==============================] - 2s 496ms/step - loss: 0.3507 - mae_euclidean: 1.3606 - jacard_coef: 0.4810 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.3299e-05
Epoch 25/200
4/4 [==============================] - 2s 497ms/step - loss: 0.3506 - mae_euclidean: 1.3292 - jacard_coef: 0.4809 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.5006e-05
Epoch 26/200
4/4 [==============================] - 2s 496ms/step - loss: 0.3414 - mae_euclidean: 1.0691 - jacard_coef: 0.4911 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 8.5901e-05
Epoch 27/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3559 - mae_euclidean: 1.3180 - jacard_coef: 0.4757 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.9553e-05
Epoch 28/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3591 - mae_euclidean: 1.6986 - jacard_coef: 0.4731 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 9.8172e-05
Epoch 29/200
4/4 [==============================] - 2s 493ms/step - loss: 0.3428 - mae_euclidean: 1.1461 - jacard_coef: 0.4895 - val_loss: 0.9995 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 2.4348e-04
Epoch 30/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3461 - mae_euclidean: 1.4201 - jacard_coef: 0.4860 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.0074e-04
Epoch 31/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3347 - mae_euclidean: 1.1866 - jacard_coef: 0.4986 - val_loss: 0.9997 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.6479e-04
Epoch 32/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3403 - mae_euclidean: 1.1343 - jacard_coef: 0.4924 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.1242e-05
Epoch 33/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3357 - mae_euclidean: 1.1377 - jacard_coef: 0.4974 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.7957e-05
Epoch 34/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3341 - mae_euclidean: 1.0978 - jacard_coef: 0.4996 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.6227e-05
Epoch 35/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3528 - mae_euclidean: 1.6263 - jacard_coef: 0.4804 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.9490e-05
Epoch 36/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3228 - mae_euclidean: 0.9696 - jacard_coef: 0.5121 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.2855e-05
Epoch 37/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3252 - mae_euclidean: 1.1013 - jacard_coef: 0.5093 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.0933e-05
Epoch 38/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3236 - mae_euclidean: 1.1611 - jacard_coef: 0.5112 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 9.2548e-05
Epoch 39/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3142 - mae_euclidean: 1.0212 - jacard_coef: 0.5219 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.2138e-04
Epoch 40/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3227 - mae_euclidean: 1.2094 - jacard_coef: 0.5122 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 7.8660e-05
Epoch 41/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3244 - mae_euclidean: 1.0948 - jacard_coef: 0.5103 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.4871e-05
Epoch 42/200
4/4 [==============================] - 2s 496ms/step - loss: 0.3244 - mae_euclidean: 1.1824 - jacard_coef: 0.5103 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 9.1089e-05
Epoch 43/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3107 - mae_euclidean: 1.0744 - jacard_coef: 0.5261 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.3223e-05
Epoch 44/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3154 - mae_euclidean: 1.0878 - jacard_coef: 0.5210 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.8086e-05
Epoch 45/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3098 - mae_euclidean: 1.1340 - jacard_coef: 0.5271 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.8710e-05
Epoch 46/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3293 - mae_euclidean: 1.2037 - jacard_coef: 0.5049 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.1277e-05
Epoch 47/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3102 - mae_euclidean: 1.0855 - jacard_coef: 0.5265 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.6131e-05
Epoch 48/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3092 - mae_euclidean: 1.1403 - jacard_coef: 0.5279 - val_loss: 0.9998 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 1.1472e-04
Epoch 49/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3078 - mae_euclidean: 1.0475 - jacard_coef: 0.5294 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 6.7359e-05
Epoch 50/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3128 - mae_euclidean: 1.0083 - jacard_coef: 0.5235 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.9633e-05
Epoch 51/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3107 - mae_euclidean: 1.0558 - jacard_coef: 0.5259 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.3958e-05
Epoch 52/200
4/4 [==============================] - 2s 496ms/step - loss: 0.3069 - mae_euclidean: 1.0413 - jacard_coef: 0.5304 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.2049e-05
Epoch 53/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3023 - mae_euclidean: 1.0936 - jacard_coef: 0.5358 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.1801e-05
Epoch 54/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2981 - mae_euclidean: 0.9954 - jacard_coef: 0.5408 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 5.0485e-05
Epoch 55/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2992 - mae_euclidean: 1.1653 - jacard_coef: 0.5396 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.9500e-05
Epoch 56/200
4/4 [==============================] - 2s 493ms/step - loss: 0.3019 - mae_euclidean: 0.9627 - jacard_coef: 0.5362 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.8101e-05
Epoch 57/200
4/4 [==============================] - 2s 494ms/step - loss: 0.3042 - mae_euclidean: 1.0287 - jacard_coef: 0.5336 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 3.7714e-05
Epoch 58/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2997 - mae_euclidean: 1.0451 - jacard_coef: 0.5390 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.0184e-05
Epoch 59/200
4/4 [==============================] - 2s 496ms/step - loss: 0.3030 - mae_euclidean: 1.1888 - jacard_coef: 0.5350 - val_loss: 1.0000 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.2897e-05
Epoch 60/200
4/4 [==============================] - 2s 498ms/step - loss: 0.3156 - mae_euclidean: 1.0978 - jacard_coef: 0.5205 - val_loss: 0.9996 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 2.2973e-04
Epoch 61/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2965 - mae_euclidean: 1.1231 - jacard_coef: 0.5427 - val_loss: 0.9999 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 8.2164e-05
Epoch 62/200
4/4 [==============================] - 2s 496ms/step - loss: 0.3040 - mae_euclidean: 1.0573 - jacard_coef: 0.5338 - val_loss: 0.9995 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 2.7040e-04
Epoch 63/200
4/4 [==============================] - 2s 495ms/step - loss: 0.3079 - mae_euclidean: 1.0688 - jacard_coef: 0.5292 - val_loss: 0.9995 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 2.5387e-04
Epoch 64/200
4/4 [==============================] - 2s 497ms/step - loss: 0.3097 - mae_euclidean: 1.1282 - jacard_coef: 0.5275 - val_loss: 0.9999 - val_mae_euclidean: 18446742974197923840.0000 - val_jacard_coef: 4.6745e-05
Epoch 65/200
4/4 [==============================] - 2s 497ms/step - loss: 0.3007 - mae_euclidean: 1.0855 - jacard_coef: 0.5377 - val_loss: 0.9996 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 2.3382e-04
Epoch 66/200
4/4 [==============================] - 2s 497ms/step - loss: 0.3102 - mae_euclidean: 1.1044 - jacard_coef: 0.5266 - val_loss: 0.9991 - val_mae_euclidean: 13835056955770535936.0000 - val_jacard_coef: 4.8740e-04
Epoch 67/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2979 - mae_euclidean: 1.0645 - jacard_coef: 0.5410 - val_loss: 0.9934 - val_mae_euclidean: 57.3900 - val_jacard_coef: 0.0033
Epoch 68/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2996 - mae_euclidean: 0.9964 - jacard_coef: 0.5390 - val_loss: 0.9986 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 7.3553e-04
Epoch 69/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2949 - mae_euclidean: 1.0318 - jacard_coef: 0.5448 - val_loss: 0.9995 - val_mae_euclidean: 11529213946556841984.0000 - val_jacard_coef: 2.5823e-04
Epoch 70/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2841 - mae_euclidean: 0.9335 - jacard_coef: 0.5576 - val_loss: 0.9998 - val_mae_euclidean: 16140899964984229888.0000 - val_jacard_coef: 1.4012e-04
Epoch 71/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2908 - mae_euclidean: 1.1867 - jacard_coef: 0.5497 - val_loss: 0.9983 - val_mae_euclidean: 4611685743549480960.0000 - val_jacard_coef: 8.7475e-04
Epoch 72/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2894 - mae_euclidean: 1.0929 - jacard_coef: 0.5514 - val_loss: 0.9957 - val_mae_euclidean: 75.8855 - val_jacard_coef: 0.0021
Epoch 73/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2873 - mae_euclidean: 0.9137 - jacard_coef: 0.5537 - val_loss: 0.9979 - val_mae_euclidean: 2305842871774740480.0000 - val_jacard_coef: 0.0011
Epoch 74/200
4/4 [==============================] - 2s 498ms/step - loss: 0.3012 - mae_euclidean: 1.3619 - jacard_coef: 0.5375 - val_loss: 0.9974 - val_mae_euclidean: 69.2515 - val_jacard_coef: 0.0013
Epoch 75/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2835 - mae_euclidean: 0.9210 - jacard_coef: 0.5583 - val_loss: 0.9957 - val_mae_euclidean: 59.7613 - val_jacard_coef: 0.0022
Epoch 76/200
4/4 [==============================] - 2s 500ms/step - loss: 0.2863 - mae_euclidean: 0.9944 - jacard_coef: 0.5551 - val_loss: 0.9978 - val_mae_euclidean: 66.0968 - val_jacard_coef: 0.0011
Epoch 77/200
4/4 [==============================] - 2s 500ms/step - loss: 0.2925 - mae_euclidean: 1.1455 - jacard_coef: 0.5482 - val_loss: 0.9890 - val_mae_euclidean: 34.2576 - val_jacard_coef: 0.0055
Epoch 78/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2917 - mae_euclidean: 1.0450 - jacard_coef: 0.5485 - val_loss: 0.9910 - val_mae_euclidean: 38.9680 - val_jacard_coef: 0.0046
Epoch 79/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2833 - mae_euclidean: 0.9998 - jacard_coef: 0.5586 - val_loss: 0.9795 - val_mae_euclidean: 26.8988 - val_jacard_coef: 0.0104
Epoch 80/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2892 - mae_euclidean: 1.0365 - jacard_coef: 0.5518 - val_loss: 0.9840 - val_mae_euclidean: 27.9931 - val_jacard_coef: 0.0081
Epoch 81/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2906 - mae_euclidean: 0.9792 - jacard_coef: 0.5500 - val_loss: 0.9926 - val_mae_euclidean: 45.6298 - val_jacard_coef: 0.0037
Epoch 82/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2817 - mae_euclidean: 1.0207 - jacard_coef: 0.5606 - val_loss: 0.9458 - val_mae_euclidean: 14.7199 - val_jacard_coef: 0.0279
Epoch 83/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2847 - mae_euclidean: 0.8814 - jacard_coef: 0.5569 - val_loss: 0.9847 - val_mae_euclidean: 42.5552 - val_jacard_coef: 0.0077
Epoch 84/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2806 - mae_euclidean: 0.9501 - jacard_coef: 0.5622 - val_loss: 0.9359 - val_mae_euclidean: 14.3443 - val_jacard_coef: 0.0331
Epoch 85/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2827 - mae_euclidean: 0.8633 - jacard_coef: 0.5592 - val_loss: 0.9501 - val_mae_euclidean: 13.9297 - val_jacard_coef: 0.0256
Epoch 86/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2842 - mae_euclidean: 1.1018 - jacard_coef: 0.5576 - val_loss: 0.9248 - val_mae_euclidean: 12.5298 - val_jacard_coef: 0.0391
Epoch 87/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2795 - mae_euclidean: 0.9200 - jacard_coef: 0.5631 - val_loss: 0.8760 - val_mae_euclidean: 10.9944 - val_jacard_coef: 0.0661
Epoch 88/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2734 - mae_euclidean: 0.9415 - jacard_coef: 0.5707 - val_loss: 0.9091 - val_mae_euclidean: 12.3863 - val_jacard_coef: 0.0476
Epoch 89/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2794 - mae_euclidean: 1.1333 - jacard_coef: 0.5638 - val_loss: 0.9015 - val_mae_euclidean: 12.4140 - val_jacard_coef: 0.0518
Epoch 90/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2768 - mae_euclidean: 0.9714 - jacard_coef: 0.5665 - val_loss: 0.8791 - val_mae_euclidean: 11.2671 - val_jacard_coef: 0.0644
Epoch 91/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2754 - mae_euclidean: 1.0188 - jacard_coef: 0.5681 - val_loss: 0.9258 - val_mae_euclidean: 13.1303 - val_jacard_coef: 0.0386
Epoch 92/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2708 - mae_euclidean: 0.8824 - jacard_coef: 0.5738 - val_loss: 0.8420 - val_mae_euclidean: 8.8745 - val_jacard_coef: 0.0858
Epoch 93/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2784 - mae_euclidean: 1.0412 - jacard_coef: 0.5647 - val_loss: 0.8792 - val_mae_euclidean: 10.5468 - val_jacard_coef: 0.0643
Epoch 94/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2731 - mae_euclidean: 0.9969 - jacard_coef: 0.5714 - val_loss: 0.8961 - val_mae_euclidean: 11.4023 - val_jacard_coef: 0.0548
Epoch 95/200
4/4 [==============================] - 2s 499ms/step - loss: 0.3032 - mae_euclidean: 1.1734 - jacard_coef: 0.5363 - val_loss: 0.7952 - val_mae_euclidean: 7.8866 - val_jacard_coef: 0.1141
Epoch 96/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2749 - mae_euclidean: 1.1841 - jacard_coef: 0.5688 - val_loss: 0.8942 - val_mae_euclidean: 10.9484 - val_jacard_coef: 0.0559
Epoch 97/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2744 - mae_euclidean: 0.9335 - jacard_coef: 0.5697 - val_loss: 0.7986 - val_mae_euclidean: 7.1317 - val_jacard_coef: 0.1120
Epoch 98/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2732 - mae_euclidean: 0.9313 - jacard_coef: 0.5710 - val_loss: 0.7639 - val_mae_euclidean: 6.5631 - val_jacard_coef: 0.1338
Epoch 99/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2712 - mae_euclidean: 1.0767 - jacard_coef: 0.5734 - val_loss: 0.6915 - val_mae_euclidean: 5.7044 - val_jacard_coef: 0.1824
Epoch 100/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2678 - mae_euclidean: 0.9197 - jacard_coef: 0.5776 - val_loss: 0.6286 - val_mae_euclidean: 5.0102 - val_jacard_coef: 0.2280
Epoch 101/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2740 - mae_euclidean: 1.2898 - jacard_coef: 0.5701 - val_loss: 0.6937 - val_mae_euclidean: 5.5913 - val_jacard_coef: 0.1809
Epoch 102/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2721 - mae_euclidean: 0.9318 - jacard_coef: 0.5724 - val_loss: 0.7198 - val_mae_euclidean: 6.2189 - val_jacard_coef: 0.1629
Epoch 103/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2656 - mae_euclidean: 0.9520 - jacard_coef: 0.5803 - val_loss: 0.6502 - val_mae_euclidean: 5.4189 - val_jacard_coef: 0.2120
Epoch 104/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2679 - mae_euclidean: 0.9776 - jacard_coef: 0.5777 - val_loss: 0.6805 - val_mae_euclidean: 5.5152 - val_jacard_coef: 0.1901
Epoch 105/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2681 - mae_euclidean: 1.0188 - jacard_coef: 0.5773 - val_loss: 0.6197 - val_mae_euclidean: 4.9735 - val_jacard_coef: 0.2348
Epoch 106/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2680 - mae_euclidean: 0.9962 - jacard_coef: 0.5773 - val_loss: 0.6184 - val_mae_euclidean: 4.2533 - val_jacard_coef: 0.2358
Epoch 107/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2652 - mae_euclidean: 1.0471 - jacard_coef: 0.5808 - val_loss: 0.6078 - val_mae_euclidean: 4.1540 - val_jacard_coef: 0.2440
Epoch 108/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2669 - mae_euclidean: 0.9487 - jacard_coef: 0.5788 - val_loss: 0.5074 - val_mae_euclidean: 3.4403 - val_jacard_coef: 0.3268
Epoch 109/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2759 - mae_euclidean: 1.0128 - jacard_coef: 0.5677 - val_loss: 0.4948 - val_mae_euclidean: 3.3481 - val_jacard_coef: 0.3380
Epoch 110/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2641 - mae_euclidean: 0.8675 - jacard_coef: 0.5824 - val_loss: 0.6221 - val_mae_euclidean: 4.7253 - val_jacard_coef: 0.2330
Epoch 111/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2579 - mae_euclidean: 0.9825 - jacard_coef: 0.5901 - val_loss: 0.5133 - val_mae_euclidean: 3.7012 - val_jacard_coef: 0.3217
Epoch 112/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2696 - mae_euclidean: 0.9456 - jacard_coef: 0.5761 - val_loss: 0.5141 - val_mae_euclidean: 3.3663 - val_jacard_coef: 0.3209
Epoch 113/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2614 - mae_euclidean: 1.0951 - jacard_coef: 0.5858 - val_loss: 0.5284 - val_mae_euclidean: 3.5039 - val_jacard_coef: 0.3086
Epoch 114/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2621 - mae_euclidean: 1.1057 - jacard_coef: 0.5851 - val_loss: 0.6109 - val_mae_euclidean: 4.4810 - val_jacard_coef: 0.2416
Epoch 115/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2586 - mae_euclidean: 0.8650 - jacard_coef: 0.5892 - val_loss: 0.5194 - val_mae_euclidean: 3.5282 - val_jacard_coef: 0.3163
Epoch 116/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2557 - mae_euclidean: 0.8813 - jacard_coef: 0.5930 - val_loss: 0.4765 - val_mae_euclidean: 3.2293 - val_jacard_coef: 0.3546
Epoch 117/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2591 - mae_euclidean: 0.8282 - jacard_coef: 0.5884 - val_loss: 0.5298 - val_mae_euclidean: 3.6721 - val_jacard_coef: 0.3074
Epoch 118/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2505 - mae_euclidean: 0.9443 - jacard_coef: 0.5995 - val_loss: 0.5121 - val_mae_euclidean: 3.6178 - val_jacard_coef: 0.3227
Epoch 119/200
4/4 [==============================] - 2s 494ms/step - loss: 0.2557 - mae_euclidean: 0.8527 - jacard_coef: 0.5928 - val_loss: 0.4868 - val_mae_euclidean: 3.3195 - val_jacard_coef: 0.3452
Epoch 120/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2495 - mae_euclidean: 0.9600 - jacard_coef: 0.6007 - val_loss: 0.4676 - val_mae_euclidean: 3.1279 - val_jacard_coef: 0.3628
Epoch 121/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2502 - mae_euclidean: 0.8585 - jacard_coef: 0.5998 - val_loss: 0.4949 - val_mae_euclidean: 3.1549 - val_jacard_coef: 0.3379
Epoch 122/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2577 - mae_euclidean: 0.9360 - jacard_coef: 0.5905 - val_loss: 0.4818 - val_mae_euclidean: 3.1001 - val_jacard_coef: 0.3497
Epoch 123/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2653 - mae_euclidean: 0.9938 - jacard_coef: 0.5818 - val_loss: 0.4603 - val_mae_euclidean: 2.9832 - val_jacard_coef: 0.3696
Epoch 124/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2529 - mae_euclidean: 0.8735 - jacard_coef: 0.5964 - val_loss: 0.4400 - val_mae_euclidean: 2.9826 - val_jacard_coef: 0.3889
Epoch 125/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2513 - mae_euclidean: 0.8053 - jacard_coef: 0.5988 - val_loss: 0.4023 - val_mae_euclidean: 2.6209 - val_jacard_coef: 0.4262
Epoch 126/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2604 - mae_euclidean: 0.9681 - jacard_coef: 0.5872 - val_loss: 0.4289 - val_mae_euclidean: 2.9010 - val_jacard_coef: 0.3997
Epoch 127/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2527 - mae_euclidean: 0.8364 - jacard_coef: 0.5967 - val_loss: 0.4859 - val_mae_euclidean: 3.2361 - val_jacard_coef: 0.3460
Epoch 128/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2527 - mae_euclidean: 1.0828 - jacard_coef: 0.5966 - val_loss: 0.3687 - val_mae_euclidean: 2.1401 - val_jacard_coef: 0.4613
Epoch 129/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2481 - mae_euclidean: 0.8220 - jacard_coef: 0.6028 - val_loss: 0.3499 - val_mae_euclidean: 2.4979 - val_jacard_coef: 0.4816
Epoch 130/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2488 - mae_euclidean: 0.7930 - jacard_coef: 0.6016 - val_loss: 0.3599 - val_mae_euclidean: 2.8534 - val_jacard_coef: 0.4707
Epoch 131/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2437 - mae_euclidean: 0.8713 - jacard_coef: 0.6081 - val_loss: 0.3942 - val_mae_euclidean: 2.8003 - val_jacard_coef: 0.4345
Epoch 132/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2609 - mae_euclidean: 0.9183 - jacard_coef: 0.5866 - val_loss: 0.3694 - val_mae_euclidean: 2.1826 - val_jacard_coef: 0.4605
Epoch 133/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2430 - mae_euclidean: 0.8960 - jacard_coef: 0.6091 - val_loss: 0.4081 - val_mae_euclidean: 2.7651 - val_jacard_coef: 0.4203
Epoch 134/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2428 - mae_euclidean: 0.7323 - jacard_coef: 0.6095 - val_loss: 0.4109 - val_mae_euclidean: 2.3370 - val_jacard_coef: 0.4176
Epoch 135/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2504 - mae_euclidean: 0.9053 - jacard_coef: 0.5998 - val_loss: 0.4249 - val_mae_euclidean: 2.9121 - val_jacard_coef: 0.4036
Epoch 136/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2446 - mae_euclidean: 0.8340 - jacard_coef: 0.6071 - val_loss: 0.3470 - val_mae_euclidean: 2.1365 - val_jacard_coef: 0.4848
Epoch 137/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2448 - mae_euclidean: 0.9199 - jacard_coef: 0.6070 - val_loss: 0.3535 - val_mae_euclidean: 2.5955 - val_jacard_coef: 0.4777
Epoch 138/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2488 - mae_euclidean: 0.9950 - jacard_coef: 0.6023 - val_loss: 0.3984 - val_mae_euclidean: 2.2473 - val_jacard_coef: 0.4302
Epoch 139/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2380 - mae_euclidean: 0.7262 - jacard_coef: 0.6156 - val_loss: 0.4623 - val_mae_euclidean: 2.9378 - val_jacard_coef: 0.3677
Epoch 140/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2375 - mae_euclidean: 0.8277 - jacard_coef: 0.6162 - val_loss: 0.3528 - val_mae_euclidean: 1.6216 - val_jacard_coef: 0.4784
Epoch 141/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2488 - mae_euclidean: 1.0794 - jacard_coef: 0.6032 - val_loss: 0.3917 - val_mae_euclidean: 2.3767 - val_jacard_coef: 0.4371
Epoch 142/200
4/4 [==============================] - 2s 500ms/step - loss: 0.2415 - mae_euclidean: 0.8987 - jacard_coef: 0.6110 - val_loss: 0.3632 - val_mae_euclidean: 2.0141 - val_jacard_coef: 0.4672
Epoch 143/200
4/4 [==============================] - 2s 501ms/step - loss: 0.2371 - mae_euclidean: 0.7727 - jacard_coef: 0.6168 - val_loss: 0.3148 - val_mae_euclidean: 1.8348 - val_jacard_coef: 0.5212
Epoch 144/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2358 - mae_euclidean: 0.7359 - jacard_coef: 0.6185 - val_loss: 0.3702 - val_mae_euclidean: 1.6764 - val_jacard_coef: 0.4597
Epoch 145/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2389 - mae_euclidean: 0.8337 - jacard_coef: 0.6144 - val_loss: 0.3598 - val_mae_euclidean: 1.5510 - val_jacard_coef: 0.4709
Epoch 146/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2365 - mae_euclidean: 0.7965 - jacard_coef: 0.6176 - val_loss: 0.3450 - val_mae_euclidean: 1.8329 - val_jacard_coef: 0.4870
Epoch 147/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2356 - mae_euclidean: 0.8028 - jacard_coef: 0.6187 - val_loss: 0.3122 - val_mae_euclidean: 1.8314 - val_jacard_coef: 0.5241
Epoch 148/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2330 - mae_euclidean: 0.7864 - jacard_coef: 0.6221 - val_loss: 0.3535 - val_mae_euclidean: 1.6760 - val_jacard_coef: 0.4777
Epoch 149/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2425 - mae_euclidean: 0.7589 - jacard_coef: 0.6098 - val_loss: 0.3402 - val_mae_euclidean: 1.5643 - val_jacard_coef: 0.4923
Epoch 150/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2350 - mae_euclidean: 1.0400 - jacard_coef: 0.6198 - val_loss: 0.3057 - val_mae_euclidean: 1.4581 - val_jacard_coef: 0.5318
Epoch 151/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2282 - mae_euclidean: 0.7669 - jacard_coef: 0.6285 - val_loss: 0.3404 - val_mae_euclidean: 1.8734 - val_jacard_coef: 0.4921
Epoch 152/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2308 - mae_euclidean: 0.8181 - jacard_coef: 0.6252 - val_loss: 0.3233 - val_mae_euclidean: 1.5433 - val_jacard_coef: 0.5114
Epoch 153/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2287 - mae_euclidean: 1.0756 - jacard_coef: 0.6278 - val_loss: 0.3565 - val_mae_euclidean: 2.0546 - val_jacard_coef: 0.4744
Epoch 154/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2306 - mae_euclidean: 0.7820 - jacard_coef: 0.6252 - val_loss: 0.3081 - val_mae_euclidean: 1.2941 - val_jacard_coef: 0.5289
Epoch 155/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2321 - mae_euclidean: 0.8691 - jacard_coef: 0.6234 - val_loss: 0.3444 - val_mae_euclidean: 1.8001 - val_jacard_coef: 0.4876
Epoch 156/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2307 - mae_euclidean: 0.7540 - jacard_coef: 0.6252 - val_loss: 0.3634 - val_mae_euclidean: 2.5841 - val_jacard_coef: 0.4669
Epoch 157/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2243 - mae_euclidean: 0.7166 - jacard_coef: 0.6337 - val_loss: 0.3242 - val_mae_euclidean: 1.2267 - val_jacard_coef: 0.5103
Epoch 158/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2351 - mae_euclidean: 0.8453 - jacard_coef: 0.6194 - val_loss: 0.3274 - val_mae_euclidean: 1.2356 - val_jacard_coef: 0.5067
Epoch 159/200
4/4 [==============================] - 2s 500ms/step - loss: 0.2207 - mae_euclidean: 0.7114 - jacard_coef: 0.6385 - val_loss: 0.3646 - val_mae_euclidean: 2.0152 - val_jacard_coef: 0.4656
Epoch 160/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2199 - mae_euclidean: 0.8186 - jacard_coef: 0.6396 - val_loss: 0.3124 - val_mae_euclidean: 1.2556 - val_jacard_coef: 0.5239
Epoch 161/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2248 - mae_euclidean: 0.7767 - jacard_coef: 0.6329 - val_loss: 0.3606 - val_mae_euclidean: 1.4555 - val_jacard_coef: 0.4699
Epoch 162/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2193 - mae_euclidean: 0.8323 - jacard_coef: 0.6405 - val_loss: 0.3475 - val_mae_euclidean: 1.3706 - val_jacard_coef: 0.4842
Epoch 163/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2190 - mae_euclidean: 0.8373 - jacard_coef: 0.6407 - val_loss: 0.3888 - val_mae_euclidean: 2.5125 - val_jacard_coef: 0.4401
Epoch 164/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2191 - mae_euclidean: 0.7207 - jacard_coef: 0.6406 - val_loss: 0.3541 - val_mae_euclidean: 1.6162 - val_jacard_coef: 0.4770
Epoch 165/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2225 - mae_euclidean: 0.7625 - jacard_coef: 0.6361 - val_loss: 0.3484 - val_mae_euclidean: 1.5998 - val_jacard_coef: 0.4833
Epoch 166/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2195 - mae_euclidean: 0.8200 - jacard_coef: 0.6401 - val_loss: 0.3270 - val_mae_euclidean: 1.2051 - val_jacard_coef: 0.5072
Epoch 167/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2224 - mae_euclidean: 0.7548 - jacard_coef: 0.6363 - val_loss: 0.3820 - val_mae_euclidean: 1.8025 - val_jacard_coef: 0.4472
Epoch 168/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2229 - mae_euclidean: 0.9876 - jacard_coef: 0.6357 - val_loss: 0.3357 - val_mae_euclidean: 1.4488 - val_jacard_coef: 0.4973
Epoch 169/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2185 - mae_euclidean: 0.6977 - jacard_coef: 0.6414 - val_loss: 0.3557 - val_mae_euclidean: 1.6661 - val_jacard_coef: 0.4753
Epoch 170/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2170 - mae_euclidean: 0.7652 - jacard_coef: 0.6434 - val_loss: 0.3169 - val_mae_euclidean: 1.1319 - val_jacard_coef: 0.5187
Epoch 171/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2178 - mae_euclidean: 0.8830 - jacard_coef: 0.6424 - val_loss: 0.3233 - val_mae_euclidean: 1.1845 - val_jacard_coef: 0.5114
Epoch 172/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2268 - mae_euclidean: 0.8841 - jacard_coef: 0.6304 - val_loss: 0.3164 - val_mae_euclidean: 1.6001 - val_jacard_coef: 0.5193
Epoch 173/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2136 - mae_euclidean: 0.8197 - jacard_coef: 0.6480 - val_loss: 0.3020 - val_mae_euclidean: 1.0686 - val_jacard_coef: 0.5361
Epoch 174/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2119 - mae_euclidean: 0.6465 - jacard_coef: 0.6504 - val_loss: 0.3141 - val_mae_euclidean: 1.2279 - val_jacard_coef: 0.5220
Epoch 175/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2101 - mae_euclidean: 0.7610 - jacard_coef: 0.6529 - val_loss: 0.3173 - val_mae_euclidean: 1.4909 - val_jacard_coef: 0.5182
Epoch 176/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2212 - mae_euclidean: 0.9298 - jacard_coef: 0.6383 - val_loss: 0.3275 - val_mae_euclidean: 1.4328 - val_jacard_coef: 0.5065
Epoch 177/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2195 - mae_euclidean: 0.7265 - jacard_coef: 0.6401 - val_loss: 0.3129 - val_mae_euclidean: 1.1077 - val_jacard_coef: 0.5234
Epoch 178/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2152 - mae_euclidean: 1.0275 - jacard_coef: 0.6459 - val_loss: 0.3369 - val_mae_euclidean: 1.5899 - val_jacard_coef: 0.4960
Epoch 179/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2118 - mae_euclidean: 0.6256 - jacard_coef: 0.6506 - val_loss: 0.3244 - val_mae_euclidean: 1.2252 - val_jacard_coef: 0.5101
Epoch 180/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2081 - mae_euclidean: 0.6659 - jacard_coef: 0.6557 - val_loss: 0.3260 - val_mae_euclidean: 1.2070 - val_jacard_coef: 0.5083
Epoch 181/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2106 - mae_euclidean: 0.7603 - jacard_coef: 0.6523 - val_loss: 0.3543 - val_mae_euclidean: 1.4435 - val_jacard_coef: 0.4768
Epoch 182/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2128 - mae_euclidean: 0.7870 - jacard_coef: 0.6493 - val_loss: 0.3213 - val_mae_euclidean: 1.4152 - val_jacard_coef: 0.5137
Epoch 183/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2111 - mae_euclidean: 0.7520 - jacard_coef: 0.6516 - val_loss: 0.3184 - val_mae_euclidean: 1.2593 - val_jacard_coef: 0.5170
Epoch 184/200
4/4 [==============================] - 2s 495ms/step - loss: 0.2044 - mae_euclidean: 0.6469 - jacard_coef: 0.6608 - val_loss: 0.3205 - val_mae_euclidean: 1.2920 - val_jacard_coef: 0.5146
Epoch 185/200
4/4 [==============================] - 2s 494ms/step - loss: 0.2001 - mae_euclidean: 0.6417 - jacard_coef: 0.6666 - val_loss: 0.3036 - val_mae_euclidean: 1.1428 - val_jacard_coef: 0.5342
Epoch 186/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2047 - mae_euclidean: 0.7207 - jacard_coef: 0.6602 - val_loss: 0.3043 - val_mae_euclidean: 1.0703 - val_jacard_coef: 0.5334
Epoch 187/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2076 - mae_euclidean: 0.8643 - jacard_coef: 0.6562 - val_loss: 0.3168 - val_mae_euclidean: 1.5778 - val_jacard_coef: 0.5188
Epoch 188/200
4/4 [==============================] - 2s 496ms/step - loss: 0.2076 - mae_euclidean: 0.7843 - jacard_coef: 0.6563 - val_loss: 0.3134 - val_mae_euclidean: 1.0872 - val_jacard_coef: 0.5227
Epoch 189/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2041 - mae_euclidean: 0.7682 - jacard_coef: 0.6611 - val_loss: 0.3294 - val_mae_euclidean: 1.5466 - val_jacard_coef: 0.5045
Epoch 190/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2012 - mae_euclidean: 0.6552 - jacard_coef: 0.6651 - val_loss: 0.3236 - val_mae_euclidean: 1.2748 - val_jacard_coef: 0.5111
Epoch 191/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1980 - mae_euclidean: 0.6610 - jacard_coef: 0.6695 - val_loss: 0.3285 - val_mae_euclidean: 1.8237 - val_jacard_coef: 0.5054
Epoch 192/200
4/4 [==============================] - 2s 497ms/step - loss: 0.2021 - mae_euclidean: 0.7040 - jacard_coef: 0.6638 - val_loss: 0.3007 - val_mae_euclidean: 0.9230 - val_jacard_coef: 0.5377
Epoch 193/200
4/4 [==============================] - 2s 497ms/step - loss: 0.1982 - mae_euclidean: 0.6551 - jacard_coef: 0.6693 - val_loss: 0.3215 - val_mae_euclidean: 1.8457 - val_jacard_coef: 0.5135
Epoch 194/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2041 - mae_euclidean: 0.9034 - jacard_coef: 0.6612 - val_loss: 0.3365 - val_mae_euclidean: 1.3517 - val_jacard_coef: 0.4964
Epoch 195/200
4/4 [==============================] - 2s 500ms/step - loss: 0.2215 - mae_euclidean: 1.0510 - jacard_coef: 0.6376 - val_loss: 0.3180 - val_mae_euclidean: 1.0438 - val_jacard_coef: 0.5174
Epoch 196/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2222 - mae_euclidean: 1.1582 - jacard_coef: 0.6367 - val_loss: 0.3129 - val_mae_euclidean: 1.2946 - val_jacard_coef: 0.5234
Epoch 197/200
4/4 [==============================] - 2s 498ms/step - loss: 0.2177 - mae_euclidean: 0.9232 - jacard_coef: 0.6426 - val_loss: 0.3541 - val_mae_euclidean: 1.7537 - val_jacard_coef: 0.4771
Epoch 198/200
4/4 [==============================] - 2s 499ms/step - loss: 0.2095 - mae_euclidean: 0.9016 - jacard_coef: 0.6538 - val_loss: 0.3921 - val_mae_euclidean: 3.0633 - val_jacard_coef: 0.4367
Epoch 199/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1990 - mae_euclidean: 0.6883 - jacard_coef: 0.6681 - val_loss: 0.3399 - val_mae_euclidean: 1.8824 - val_jacard_coef: 0.4926
Epoch 200/200
4/4 [==============================] - 2s 498ms/step - loss: 0.1982 - mae_euclidean: 0.7672 - jacard_coef: 0.6692 - val_loss: 0.3216 - val_mae_euclidean: 1.0686 - val_jacard_coef: 0.5134
Model trained for 395.8336317539215s

plot_history(history_sa_unet4, "SA Unet4")

Comparing the graphs from model_sa_unet4 and model_sa_unet2 gives indication for the effect of different beta_1 and beta_2. All the differences are within the noise of the curves. It would be preferred to use the default beta_1 and beta_2.

Final model selection

The model model_sa_unet1 will be accepted as final. Motivation:

• Best jacard coefficient; val_mae_euclidean among the best
• no overfitting
• light / fast model architecture with only 838279 parameters
• loss function experiments indicated dice loss as more suitable

Key implementation details:

• block_size=19, keep_prob=0.8, start_neurons=20
• Adam optimizer with learning rate 1e-2, beta_1=0.9, beta_2=0.999
• dice coefficient loss is used
• trained for 200 epochs
• batch size is 8

The trained model can be loaded from the file "2022-02-17 SA-UNet1 200epochs.hdf5". The custom metrics have to be specified during loading.

# best_model = model_sa_unet1
best_model = tf.keras.models.load_model('2022-02-17 SA-UNet1 200epochs.hdf5',
                                           custom_objects={'dice_coef_loss': dice_coef_loss,
                                                          'jacard_coef': jacard_coef,
                                                          'mae_euclidean': mae_euclidean,
                                                          'DropBlock2D' : DropBlock2D}
                                           )

Final model analysis

best_model.summary()

Model: "SA_UNet"
__________________________________________________________________________________________________
 Layer (type)                   Output Shape         Param #     Connected to                     
==================================================================================================
 input_1 (InputLayer)           [(None, 176, 256, 1  0           []                               
                                )]                                                                
                                                                                                  
 conv2d (Conv2D)                (None, 176, 256, 20  200         ['input_1[0][0]']                
                                )                                                                 
                                                                                                  
 drop_block2d (DropBlock2D)     (None, 176, 256, 20  0           ['conv2d[0][0]']                 
                                )                                                                 
                                                                                                  
 batch_normalization (BatchNorm  (None, 176, 256, 20  80         ['drop_block2d[0][0]']           
 alization)                     )                                                                 
                                                                                                  
 activation (Activation)        (None, 176, 256, 20  0           ['batch_normalization[0][0]']    
                                )                                                                 
                                                                                                  
 conv2d_1 (Conv2D)              (None, 176, 256, 20  3620        ['activation[0][0]']             
                                )                                                                 
                                                                                                  
 drop_block2d_1 (DropBlock2D)   (None, 176, 256, 20  0           ['conv2d_1[0][0]']               
                                )                                                                 
                                                                                                  
 batch_normalization_1 (BatchNo  (None, 176, 256, 20  80         ['drop_block2d_1[0][0]']         
 rmalization)                   )                                                                 
                                                                                                  
 activation_1 (Activation)      (None, 176, 256, 20  0           ['batch_normalization_1[0][0]']  
                                )                                                                 
                                                                                                  
 max_pooling2d (MaxPooling2D)   (None, 88, 128, 20)  0           ['activation_1[0][0]']           
                                                                                                  
 conv2d_2 (Conv2D)              (None, 88, 128, 40)  7240        ['max_pooling2d[0][0]']          
                                                                                                  
 drop_block2d_2 (DropBlock2D)   (None, 88, 128, 40)  0           ['conv2d_2[0][0]']               
                                                                                                  
 batch_normalization_2 (BatchNo  (None, 88, 128, 40)  160        ['drop_block2d_2[0][0]']         
 rmalization)                                                                                     
                                                                                                  
 activation_2 (Activation)      (None, 88, 128, 40)  0           ['batch_normalization_2[0][0]']  
                                                                                                  
 conv2d_3 (Conv2D)              (None, 88, 128, 40)  14440       ['activation_2[0][0]']           
                                                                                                  
 drop_block2d_3 (DropBlock2D)   (None, 88, 128, 40)  0           ['conv2d_3[0][0]']               
                                                                                                  
 batch_normalization_3 (BatchNo  (None, 88, 128, 40)  160        ['drop_block2d_3[0][0]']         
 rmalization)                                                                                     
                                                                                                  
 activation_3 (Activation)      (None, 88, 128, 40)  0           ['batch_normalization_3[0][0]']  
                                                                                                  
 max_pooling2d_1 (MaxPooling2D)  (None, 44, 64, 40)  0           ['activation_3[0][0]']           
                                                                                                  
 conv2d_4 (Conv2D)              (None, 44, 64, 80)   28880       ['max_pooling2d_1[0][0]']        
                                                                                                  
 drop_block2d_4 (DropBlock2D)   (None, 44, 64, 80)   0           ['conv2d_4[0][0]']               
                                                                                                  
 batch_normalization_4 (BatchNo  (None, 44, 64, 80)  320         ['drop_block2d_4[0][0]']         
 rmalization)                                                                                     
                                                                                                  
 activation_4 (Activation)      (None, 44, 64, 80)   0           ['batch_normalization_4[0][0]']  
                                                                                                  
 conv2d_5 (Conv2D)              (None, 44, 64, 80)   57680       ['activation_4[0][0]']           
                                                                                                  
 drop_block2d_5 (DropBlock2D)   (None, 44, 64, 80)   0           ['conv2d_5[0][0]']               
                                                                                                  
 batch_normalization_5 (BatchNo  (None, 44, 64, 80)  320         ['drop_block2d_5[0][0]']         
 rmalization)                                                                                     
                                                                                                  
 activation_5 (Activation)      (None, 44, 64, 80)   0           ['batch_normalization_5[0][0]']  
                                                                                                  
 max_pooling2d_2 (MaxPooling2D)  (None, 22, 32, 80)  0           ['activation_5[0][0]']           
                                                                                                  
 conv2d_6 (Conv2D)              (None, 22, 32, 160)  115360      ['max_pooling2d_2[0][0]']        
                                                                                                  
 drop_block2d_6 (DropBlock2D)   (None, 22, 32, 160)  0           ['conv2d_6[0][0]']               
                                                                                                  
 batch_normalization_6 (BatchNo  (None, 22, 32, 160)  640        ['drop_block2d_6[0][0]']         
 rmalization)                                                                                     
                                                                                                  
 activation_6 (Activation)      (None, 22, 32, 160)  0           ['batch_normalization_6[0][0]']  
                                                                                                  
 lambda (Lambda)                (None, 22, 32, 1)    0           ['activation_6[0][0]']           
                                                                                                  
 lambda_1 (Lambda)              (None, 22, 32, 1)    0           ['activation_6[0][0]']           
                                                                                                  
 concatenate (Concatenate)      (None, 22, 32, 2)    0           ['lambda[0][0]',                 
                                                                  'lambda_1[0][0]']               
                                                                                                  
 conv2d_7 (Conv2D)              (None, 22, 32, 1)    98          ['concatenate[0][0]']            
                                                                                                  
 multiply (Multiply)            (None, 22, 32, 160)  0           ['activation_6[0][0]',           
                                                                  'conv2d_7[0][0]']               
                                                                                                  
 conv2d_8 (Conv2D)              (None, 22, 32, 160)  230560      ['multiply[0][0]']               
                                                                                                  
 drop_block2d_7 (DropBlock2D)   (None, 22, 32, 160)  0           ['conv2d_8[0][0]']               
                                                                                                  
 batch_normalization_7 (BatchNo  (None, 22, 32, 160)  640        ['drop_block2d_7[0][0]']         
 rmalization)                                                                                     
                                                                                                  
 activation_7 (Activation)      (None, 22, 32, 160)  0           ['batch_normalization_7[0][0]']  
                                                                                                  
 conv2d_transpose (Conv2DTransp  (None, 44, 64, 80)  115280      ['activation_7[0][0]']           
 ose)                                                                                             
                                                                                                  
 concatenate_1 (Concatenate)    (None, 44, 64, 160)  0           ['conv2d_transpose[0][0]',       
                                                                  'activation_5[0][0]']           
                                                                                                  
 conv2d_9 (Conv2D)              (None, 44, 64, 80)   115280      ['concatenate_1[0][0]']          
                                                                                                  
 drop_block2d_8 (DropBlock2D)   (None, 44, 64, 80)   0           ['conv2d_9[0][0]']               
                                                                                                  
 batch_normalization_8 (BatchNo  (None, 44, 64, 80)  320         ['drop_block2d_8[0][0]']         
 rmalization)                                                                                     
                                                                                                  
 activation_8 (Activation)      (None, 44, 64, 80)   0           ['batch_normalization_8[0][0]']  
                                                                                                  
 conv2d_10 (Conv2D)             (None, 44, 64, 80)   57680       ['activation_8[0][0]']           
                                                                                                  
 drop_block2d_9 (DropBlock2D)   (None, 44, 64, 80)   0           ['conv2d_10[0][0]']              
                                                                                                  
 batch_normalization_9 (BatchNo  (None, 44, 64, 80)  320         ['drop_block2d_9[0][0]']         
 rmalization)                                                                                     
                                                                                                  
 activation_9 (Activation)      (None, 44, 64, 80)   0           ['batch_normalization_9[0][0]']  
                                                                                                  
 conv2d_transpose_1 (Conv2DTran  (None, 88, 128, 40)  28840      ['activation_9[0][0]']           
 spose)                                                                                           
                                                                                                  
 concatenate_2 (Concatenate)    (None, 88, 128, 80)  0           ['conv2d_transpose_1[0][0]',     
                                                                  'activation_3[0][0]']           
                                                                                                  
 conv2d_11 (Conv2D)             (None, 88, 128, 40)  28840       ['concatenate_2[0][0]']          
                                                                                                  
 drop_block2d_10 (DropBlock2D)  (None, 88, 128, 40)  0           ['conv2d_11[0][0]']              
                                                                                                  
 batch_normalization_10 (BatchN  (None, 88, 128, 40)  160        ['drop_block2d_10[0][0]']        
 ormalization)                                                                                    
                                                                                                  
 activation_10 (Activation)     (None, 88, 128, 40)  0           ['batch_normalization_10[0][0]'] 
                                                                                                  
 conv2d_12 (Conv2D)             (None, 88, 128, 40)  14440       ['activation_10[0][0]']          
                                                                                                  
 drop_block2d_11 (DropBlock2D)  (None, 88, 128, 40)  0           ['conv2d_12[0][0]']              
                                                                                                  
 batch_normalization_11 (BatchN  (None, 88, 128, 40)  160        ['drop_block2d_11[0][0]']        
 ormalization)                                                                                    
                                                                                                  
 activation_11 (Activation)     (None, 88, 128, 40)  0           ['batch_normalization_11[0][0]'] 
                                                                                                  
 conv2d_transpose_2 (Conv2DTran  (None, 176, 256, 20  7220       ['activation_11[0][0]']          
 spose)                         )                                                                 
                                                                                                  
 concatenate_3 (Concatenate)    (None, 176, 256, 40  0           ['conv2d_transpose_2[0][0]',     
                                )                                 'activation_1[0][0]']           
                                                                                                  
 conv2d_13 (Conv2D)             (None, 176, 256, 20  7220        ['concatenate_3[0][0]']          
                                )                                                                 
                                                                                                  
 drop_block2d_12 (DropBlock2D)  (None, 176, 256, 20  0           ['conv2d_13[0][0]']              
                                )                                                                 
                                                                                                  
 batch_normalization_12 (BatchN  (None, 176, 256, 20  80         ['drop_block2d_12[0][0]']        
 ormalization)                  )                                                                 
                                                                                                  
 activation_12 (Activation)     (None, 176, 256, 20  0           ['batch_normalization_12[0][0]'] 
                                )                                                                 
                                                                                                  
 conv2d_14 (Conv2D)             (None, 176, 256, 20  3620        ['activation_12[0][0]']          
                                )                                                                 
                                                                                                  
 drop_block2d_13 (DropBlock2D)  (None, 176, 256, 20  0           ['conv2d_14[0][0]']              
                                )                                                                 
                                                                                                  
 batch_normalization_13 (BatchN  (None, 176, 256, 20  80         ['drop_block2d_13[0][0]']        
 ormalization)                  )                                                                 
                                                                                                  
 activation_13 (Activation)     (None, 176, 256, 20  0           ['batch_normalization_13[0][0]'] 
                                )                                                                 
                                                                                                  
 conv2d_15 (Conv2D)             (None, 176, 256, 1)  21          ['activation_13[0][0]']          
                                                                                                  
 activation_14 (Activation)     (None, 176, 256, 1)  0           ['conv2d_15[0][0]']              
                                                                                                  
==================================================================================================
Total params: 840,039
Trainable params: 838,279
Non-trainable params: 1,760
__________________________________________________________________________________________________

# photos with 500x scale
plot_train_val(best_model, 1, 1)

# photos with 1000x scale
plot_train_val(best_model, 16, 5)

# plotting histogram of a valdiation picture
idx = 1
pred = best_model.predict(np.expand_dims(X_val[idx], 0)) 
pred = np.squeeze(pred)
gt_pixels = y_val[idx]
gt_pixels = np.squeeze(gt_pixels)
border = pred[gt_pixels == 1]
grain = pred[gt_pixels == 0]
plt.hist(border, bins=20, density=False, label='true borders', histtype='step')
plt.hist(grain, bins=20, density=False, label='true grains', histtype='step')
plt.title("histogram of border probability")
plt.ylabel("number of pixels in bin")
plt.yscale('log')
plt.grid()
plt.legend()
plt.show()

The histograms shows good separation between border and grain pixels. The accepted threshold of 0.5 is good, reasonable changes of the threshold will not yield very different results. Looking at the histogram only, one could notice the classifficator performance is quite poor. However, when the spatial acceptance is added from the picture above, the results look quite promising. It appears that often FP and FN pixel borders are running in parallel - the shift could be due to errors in the ground truth. Some FP are actually quite questionable for the human and some people would classify them as borders.

# view the distance maps 
idx = 1
y_pred = best_model.predict(np.expand_dims(X_val[idx], 0))[0]
y_true = y_val[idx]

y_true_inv = (y_true == 0)
y_true_inv = tf.cast(y_true_inv, tf.uint8)
y_true_distance_map = tfa.image.euclidean_dist_transform(y_true_inv)  

y_pred_inv = (y_pred < 0.5)
y_pred_inv = tf.cast(y_pred_inv, tf.uint8)
y_pred_distance_map = tfa.image.euclidean_dist_transform(y_pred_inv)

error_euc_distances = tf.math.abs(tf.math.subtract(y_pred_distance_map, y_true_distance_map))
error_euc_distances = tf.squeeze(error_euc_distances).numpy()

plt.figure(figsize=(16, 7))
plt.subplot(1, 2, 1)
im = plt.imshow(y_true_distance_map, cmap='jet', vmax=40)
plt.colorbar(im, shrink=0.8)
plt.title("True euclidean distances")
plt.subplot(1, 2, 2)
im = plt.imshow(y_pred_distance_map, cmap='jet', vmax=40)
plt.colorbar(im, shrink=0.8)
plt.title("Predicted euclidean distances")
plt.show()

plt.figure(figsize=(12, 7))
im = plt.imshow(error_euc_distances, cmap='jet')
plt.colorbar(im)
plt.title("Absolute error in euclidean distances")
plt.show()

On the picture above, the zones with biggest error are in red.

Final model evaluation

# model evaluation on train set
best_model.evaluate(X_train, y_train, batch_size=8)

4/4 [==============================] - 1s 58ms/step - loss: 0.2292 - mae_euclidean: 0.7226 - jacard_coef: 0.6282

[0.22923870384693146, 0.7226119041442871, 0.6281639337539673]

# model evaluation on validation set
best_model.evaluate(X_val, y_val, batch_size=8)

1/1 [==============================] - 0s 62ms/step - loss: 0.2979 - mae_euclidean: 1.2925 - jacard_coef: 0.5409

[0.2979312539100647, 1.292516827583313, 0.540920078754425]

# model evaluation on test set
best_model.evaluate(X_test, y_test, batch_size=8)

1/1 [==============================] - 0s 328ms/step - loss: 0.3015 - mae_euclidean: 1.1366 - jacard_coef: 0.5367

[0.3015058636665344, 1.1366015672683716, 0.5366954803466797]

The model performance is similar on the test and the validation dataset.

# analyzing all test images (original and augmented by median filter)
plt.figure(figsize=(16, 12))

for idx in range(X_test.shape[0]):
    im = visual_eval(model=best_model, test_img=X_test[idx], mask_expected=y_test[idx], show=False)
    plt.subplot(2, 2, idx+1)
    plt.imshow(im)
    plt.title(f"Test image {idx}")
plt.tight_layout()
plt.show()

Probably the FN pixels could be improved by adding more training data.

Running the model on production images

(no manual labeling available)

Functions needed for production

# functions needed to analyze full picture 1280x880 pixels
def create_patch_tensor(image):
    """
    prepares patches from image.
    the output tensor is with shape [25, 176, 256, 1] - 25 patches with resolution 176x256
    """
    im = np.expand_dims([image], -1)
    p = tf.image.extract_patches(images=im,
                           sizes=[1, 176, 256, 1],
                           strides=[1, 176, 256, 1],
                           rates=[1, 1, 1, 1],
                           padding='VALID') # small size of the image is lost
    n, r, c = p.shape[0], p.shape[1], p.shape[2]
    if r != 5 or c !=5:
        print(f"The function is intended for input shape of 880-893x1280 but shape {image.shape} was given")
    p = tf.reshape(p, [n*r*c, 176, 256, 1])
    return p


def get_border_mask_and_preview(img, model=best_model, thresh=0.50):
    assert (img.shape == (880, 1280))
    img = img_as_float32(img)
    img_p = create_patch_tensor(img)
    total_preview = np.zeros(tf.repeat(img_p, 3, axis=-1).shape)
    total_mask = np.zeros(tf.squeeze(img_p).shape, dtype=np.uint8)

    for i in range(5*5):
        patch = img_p[i]
        patch = img_as_float32(patch)
        preview = patch.copy()
        preview = tf.repeat(preview, 3, -1).numpy()
      
        pred = model.predict(np.expand_dims(patch, 0)) 
        
        if thresh !=0.5:
            mask = np.zeros(patch.shape[0:2])
            mask[np.squeeze(pred) > thresh] = 1
            preview[mask == 1] = (0.5, 0.5, 0.0)
        
        mask = np.zeros(patch.shape[0:2])
        mask[np.squeeze(pred) > 0.50] = 1
        preview[mask == 1] = (1.0, 0.0, 0.0)
                
        total_mask[i] = mask
        total_preview[i] = preview
        
    total_mask = np.reshape(total_mask, (5, 5, IMG_HEIGHT, IMG_WIDTH))
    total_mask = np.hstack(np.hstack(total_mask))
    
    total_preview = np.reshape(total_preview, (5, 5, IMG_HEIGHT, IMG_WIDTH, 3))
    total_preview = np.hstack(np.hstack(total_preview))

    return total_mask, total_preview


def get_grain_prop_and_preview(img, mask_boundaries):
    """
    img - input image, grayscale
    mask_boundaries - mask for the boundaries
    returns:
        - grain statistics
        - colored preview of the grains
    uses waterhed algorithm
    """
    if img.ndim != 2:
        raise ValueError("Grayscale image expected")
    if img.dtype != np.float64 and img.dtype != np.float32:
        raise ValueError("Image must be float32 or float64!")

    if img.dtype == np.float64:
        img = img_as_float32(img)

    # obtainning distance map and peaks
    mask_boundaries = mask_boundaries.astype(bool)
    mask_grains = ~mask_boundaries
    distance_map = ndimage.distance_transform_edt(mask_grains)
#     plt.imshow(mask_grains, cmap='jet')
#     return 0, 0
    peak_idx = peak_local_max(distance_map, min_distance=5, exclude_border=True)

    # filter unique peaks only, that are not connected to other peaks
    dmap = distance_map.copy() # dmap will be modified by flood fill
    unique_idx = []
    for peak in peak_idx: # start from biggest peaks to the smallest
        max_dist = dmap[peak[0], peak[1]] # the value of the peak at x, y
        if max_dist > 0:
            # this peak has not been eaten so far
            unique_idx.append(peak)
            dmap = flood_fill(dmap, (peak[0], peak[1]), new_value=0, tolerance=0.6*max_dist) #tolerance=0.3*max_dist
    unique_idx = np.array(unique_idx)

    # mask for major peaks that survived the flood_fill
    peak_mask = np.zeros_like(distance_map, dtype=bool)
    peak_mask[tuple(unique_idx.T)] = True
    cross = np.array([[0, 1, 0],
                     [1, 1, 1],
                     [0, 1, 0],])
    peak_preview = binary_dilation(peak_mask, cross)
    
    # mask for all peaks
    peak_mask_all = np.zeros_like(distance_map, dtype=bool)
    peak_mask_all[tuple(peak_idx.T)] = True
    peak_preview_all = binary_dilation(peak_mask_all, cross)

    # Perform connected component analysis then apply Watershed
    markers = ndimage.label(peak_mask, structure=np.ones((3, 3)))[0]
    labels = watershed(-distance_map, markers, mask=mask_grains, watershed_line=True)
    labels = clear_border(labels)
    peak_preview[labels == 0] = 0 # remove the center annotation
    
    # create statistics
    regions = measure.regionprops(labels)  
    grain_prop = {'area': list(),
                 'd_max' : list()}
    for prop in regions:
        area = prop.filled_area
        perimeter = prop.perimeter
        grain_prop['area'].append(area + 0.5 * perimeter) 
        grain_prop['d_max'].append(prop.feret_diameter_max)
        
    # create colored picture
    x = np.linspace(0.0, 1, 20)
    colors = plt.get_cmap('tab20')(x)[:,:3] 
    colors = np.delete(colors, [14, 15], 0) # remove grays

    grain_preview = color.label2rgb(labels, image=img, colors=colors, bg_label=0, alpha=0.2)
    # grain_preview[peak_preview_all] = (0.5, 0.5, 0.5) # show minor centers of grains
    grain_preview[peak_preview] = (1, 1, 1) # show centers of grains
    grain_preview[~mask_grains] = (0.3, 0, 0) # show borders from the ML

    return grain_prop, grain_preview


def detec_grains_in_folder(path, model=best_model):
    """
    path should be without path separators
    makes preview and statistics for a folder
    """
    path_out = path

    df_folder = pd.DataFrame()
    for image in os.listdir(path):
        if image.split(".")[-1] == 'jpg':
            img_path = path + "\\" + image
            print(f"Working on {img_path}")
    
            # reading and processing the image
            img = io.imread(img_path, as_gray=True)
            img = img_as_float32(img)
 
            mask, mask_preview = get_border_mask_and_preview(img, model)
            grain_prop, colored_preview = get_grain_prop_and_preview(img, mask)

            # saving the ML borders
            vis = mask_preview.copy()
            vis[vis >1] = 1
            vis = img_as_ubyte(vis)
            new_filename = img_path.replace(".jpg", "_borders.tif")
            io.imsave(new_filename, vis)

            # saving colored grains
            vis = colored_preview.copy()
            vis[vis >1] = 1
            vis = img_as_ubyte(vis)
            new_filename = img_path.replace(".jpg", "_grains.tif")
            io.imsave(new_filename, vis)
            
            df_image = pd.DataFrame(grain_prop)
            df_image.area = df_image.area * 1 # number of pixels
            df_image.d_max = df_image.d_max * 1 # pixels
            df_image["file"] = image
            
            df_folder = pd.concat([df_folder, df_image], axis=0, ignore_index=True)
            
    df_folder.to_csv(path_out + "\\" + "grain_summary.csv")
        
    return df_folder

Production run

# creating summaries in each folder
# grain_stats = detec_grains_in_folder(r"datasets\production") # already done

The processed images are stored in the original folder.

img = io.imread(r"datasets\production\production_1_borders.tif")
plt.figure(figsize=(14,10))
plt.imshow(img)
plt.title('example border preview')
plt.show()

Backup

# removing image footer and renaming
# path = r"datasets\production"
# i = 1
# for image in os.listdir(path):
#     if image.split(".")[-1] == 'jpg':
#         img_path = path + "\\" + image
#         img = io.imread(img_path, as_gray=True)
#         img = img_as_ubyte(img)
#         img =  img[2:882,:]
#         io.imsave(path + f"\\production_{i}.jpg", img)
#         i += 1

Discussion

Comparison to the ML approach using XGBoost https://github.com/Chehlarov/Machine-Learning/tree/main/00%20-%20project

• The ML approach required border pixel marking of 3 pixels to achieve good results. With Unets, the bordering labeling is single pixel and good results are obtained.
• The Unet implementation does not require any preprocessing. The ML approach required denoising and histogram matching.
• The Unet does not require postprocessing like skeletonization or morphological operations.
• The Unet was trained simultaneously on patches with two different scales (x500 and x1000). The model performs well on both scale levels.

Notes on the dataset:

• Shadows on the image have dominant direction and create false bias. One could apply augmentation by rotation/mirroring. Neither rotations nor mirroring are applied here, because photos are taken on the same SEM machine and the shadow direction is always the same.

Lessons learned:

• The much smaller SA-Unet performed better than larger Unets. The smaller network, and dropblock prevent overfitting effectively.
• Be careful with augmentations in the color space domain. Some augmentation might produce pictures impossible to be separated by human. If applying such approach, create the augmented pictures and manually inspect the feasibility of their use.

Ideas for improvement:

• Add loss function that does not put big penalty on false positives close to GT positive predictions. The manual pixel by pixel labeling cannot be 100% correct, such loss function might work better.
• Add weighted loss, obtained from distance map, based on GT. See "Distance Map Loss Penalty Term for Semantic Segmentation".
• Augmentation by applying elastic deformations could be quite effective - https://www.kaggle.com/ori226/data-augmentation-with-elastic-deformations
• Train on images with more scale levels, if such are going to be used. Augmentation could also be used for down scaling.
• If more training data is to be added, it is best to apply the model on images and look for image types where the model under-performs. Manually correct a couple of patches, no need to correct the full picture. It is better to spend the labeling resources by covering several images by 1 or 2 patches. The added value of labeling all patches in an image is minimal, because the local features repeat a lot in one image.

Conclusions:

• Impressive results were achieved by training on patches of two labeled images.
• Small (trainable parameters: 838279) and effective model is proposed, ready to be applied on production scale.

References

• Spatial Attention U-Net: https://arxiv.org/abs/2004.03696, https://github.com/clguo/SA-UNet
• U-net: https://arxiv.org/abs/1505.04597
• Attention U-net: https://arxiv.org/pdf/1804.03999.pdf
• Attention Residual U-Net: https://arxiv.org/abs/2004.03702
• code used from: https://youtu.be/L5iV5BHkMzM
• DropBlock: A regularization method for convolutional networks: https://arxiv.org/pdf/1810.12890.pdf
• A survey of loss functions for semantic segmentation https://arxiv.org/pdf/2006.14822.pdf
• Distance Map Loss Penalty Term for Semantic Segmentation https://arxiv.org/pdf/1908.03679.pdf